Is “Unity Gain” Still the #1 Objective?

Maybe YES!

A Technical Paper prepared for SCTE•ISBE by

John Ulm
Engineering Fellow, Broadband Systems
CommScope – CTO Network Solutions team
Moultonborough, NH 03254
+1 (978) 609-6028
[email protected]

Dr. Zoran Maricevic

Engineering Fellow
CommScope – CTO Network Solutions team
15 Sterling Drive Wallingford, CT 06117
+1 (203) 303-6547
[email protected]

Dr. Frank O’Keeffe

Distinguished Systems Engineer
CommScope – CTO Network Solutions team
2 Johnstown Close, Glounthaune, Cork, Ireland T45V094
(+353) 87 668 8789
[email protected]

© 2020 SCTE•ISBE and NCTA. All rights reserved. 1

 Table of Contents
Title Page Number
1. Introduction .......................................................................................................................................... 4
2. Unity Gain Overview ........................................................................................................................... 4
3. Network Capacity Planning for Cable 10G ......................................................................................... 5
3.1. The “Basic” Traffic Engineering Formula ............................................................................... 6
3.2. Broadband Subscriber Traffic Consumption .......................................................................... 7
3.3. Network Capacity Modeling for Cable 10G Downstream....................................................... 8
3.4. Network Capacity Planning for DOCSIS 4.0 ESD ................................................................. 9
3.4.1. Cable 10G Service Tiers ........................................................................................ 9
3.4.2. DOCSIS 4.0 ESD – Tavg and Nsub targets ........................................................ 10
3.4.3. DOCSIS 4.0 ESD – Network Capacity Requirements ......................................... 11
4. Simulation Model Overview ............................................................................................................... 11
4.1. Cable Model – PSD and TCP .............................................................................................. 11
4.2. Amplifier Model..................................................................................................................... 11
4.3. System Noise Performance ................................................................................................. 13
4.4. Amplifier Spacing – Typical and Stretch Networks .............................................................. 13
5. Conventional Wisdom Using Unity Gain Approach ........................................................................... 14
6. Game Changer – DOCSIS Compensates When Unity Gain Lags ................................................... 17
6.1. Estimating HP/Subscriber Distribution across N+X HFC Plant............................................ 20
6.2. DOCSIS Optimized Network Capacity Analysis .................................................................. 21
6.3. Impact of shorter drop cables (66’/100’/150’) ....................................................................... 22
6.4. Impact of Super-stretched Links (56dB) .............................................................................. 24
6.5. Downstream ESD Capacity for various US/DS splits .......................................................... 26
6.6. Comparing 108-1218MHz FDX vs. 606-1794MHz ESD DS Capacity ................................. 29
6.7. Improving on the DOCSIS Weighted Average Capacity ...................................................... 29
6.8. Handling the Corner Cases – Meeting Tmax Burst QoE ..................................................... 31
7. Cost and Logistic Analysis ................................................................................................................ 33
8. Conclusion ......................................................................................................................................... 36
Acknowledgements ..................................................................................................................................... 37
Bibliography & References.......................................................................................................................... 38
Abbreviations .............................................................................................................................................. 39

List of Figures
Title Page Number
Figure 1 – A big picture view of a Hybrid Fiber Coaxial (HFC) network ....................................................... 4
Figure 2 – Illustration of the "Unity Gain" concept, applied to an RF amplifier cascade............................... 5
Figure 3 – Downstream Average Bandwidth per Subscriber through Jan ‘20 .............................................. 7
Figure 4 – Upstream Average Bandwidth per Subscriber through Jan ‘20 .................................................. 7
Figure 5 – 1218/204 MHz System – Subs per SG, DOCSIS Spectrum Needs ............................................ 8
Figure 6 – 1218/204 MHz System – DOCSIS Usage: Tmax, Tavg, IP Video .............................................. 9
Figure 7 – PSD Profile Showing Tilt/Drop/Tilt Characteristic ...................................................................... 12
Figure 8 – Inter-amplifier Span Attenuation & Amp Gain (With 44 dB Gain limit) in a Stretched Plant ...... 12
Figure 9 – Amplifier Spacing – Single Output Line Extenders (LE) ............................................................ 13

© 2020 SCTE•ISBE and NCTA. All rights reserved. 2

Figure 10 – Amplifier Spacing – Multi-Output Bridgers (MB) ...................................................................... 14
Figure 11 – End-of-Line Throughput - 108-1218 MHz, 150' Drop cables ................................................... 15
Figure 12 – Stretch 108-1218 MHz Plant Bit-loading – Tap 5 for N+2 and N+5 ........................................ 15
Figure 13 – End-of-Line Throughput - 492-1794 MHz, 150' Drop cables ................................................... 16
Figure 14 – Typical 492-1794 MHz Plant Bit-loading – Tap 5 for N+2 and N+5 ........................................ 16
Figure 15 – Stretch 492-1794 MHz Plant Bit-loading – Tap 5 for N+2 and N+5 ........................................ 17
Figure 16 – 108-1218 MHz Net Throughput vs. Tap Position in N+6 Plant, 150' Drop cables ................. 18
Figure 17 – 492-1794 MHz Throughput vs. Tap Position, N+6 Typical Plant, 150' Drops ........................ 18
Figure 18 – 492-1794 MHz Throughput vs. Tap Position, N+6 Stretch Plant, 150' Drops ........................ 19
Figure 19 – End-of-Line vs. Weighted Avg Throughput - 108-1218 MHz, 150' Drop cables...................... 21
Figure 20 – End-of-Line vs. Weighted Avg Throughput - 492-1794 MHz, 150' Drop cables...................... 22
Figure 21 – Net Throughput vs Tap Position in N+6 Typical Plant, 66/100/150 ft drop cable .................... 23
Figure 22 – Net Throughput vs Tap Position in N+6 Stretch Plant, 66/100/150 ft drop cable .................... 23
Figure 23 – 492-1794 Typical Plant with 1 or 2 Super-stretch 56 dB Links ................................................ 25
Figure 24 –Weighted Average DS Capacity with 1 or 2 Super-stretch 56 dB Links ................................... 26
Figure 25 – Net Throughput vs. Tap Position in N+6 Typical Plant, different US/DS splits ....................... 27
Figure 26 – Weighted Average DS Capacity for different US/DS splits ..................................................... 28
Figure 27 – Weighted Avg Throughput - 108-1218 MHz vs. 606-1794 MHz, 150' Drop cables ................ 29
Figure 28 – Example Bit-loading for Near and Far Homes ......................................................................... 30
Figure 29 – Example Channel Capacities for various Tap Positions .......................................................... 30
Figure 30 – I-CCAP N+4 HFC network with 42/54 MHz sub-split .............................................................. 33
Figure 31 – CAPEX estimates for 1218/204 MHz N+4 (both without and with new tap faceplates), 1218
MHz FDX N+0, 1794 MHz ESD N+4 upgrades .................................................................................. 34
Figure 32 – DAA N+0 HFC Network Model for 1.2 GHz FDX Upgrade...................................................... 34
Figure 33 – DAA N+4 HFC Network Model for 1.8 GHz ESD upgrade ...................................................... 34
Figure 34 – Monte-Carlo analysis of “premium” for 1.8 GHz ESD case ..................................................... 35
Figure 35 – EoL vs. Weighted Avg Throughput – Typical and Stretch Plants ............................................ 37

List of Tables
Title Page Number
Table 1 – Summary of SLA Options for 10G PON & 1218 MHz Plants ..................................................... 10
Table 2 – Example RF Amps and Homes Passed per Node for N+X ........................................................ 21
Table 3 – Summary of SLA Options for 1794 MHz Plants .......................................................................... 28
Table 4 – Estimate of Labor force required to replace amps and taps in 5 years ...................................... 36

© 2020 SCTE•ISBE and NCTA. All rights reserved. 3

1. Introduction
As of April 2020, all the 1,000+ pages of the Data Over Cable Service Interface Specification (DOCSIS)
4.0 PHY and MULPI specs, combined, are out and about. That’s step one in making the 1,794 MHz top
frequency in the downstream, combined with up to 684 MHz for the top frequency in the upstream, a new
reality for the hybrid fiber-coax (HFC) plants. As part of the greater 10G initiative, this is what will be
needed to support the many Gbps in both the upstream and the downstream for nodes with amplifier
cascades (a.k.a. N+X). The multiple system operators’ (MSOs’) successful business model and existing
HFC broadband networks are our starting point. All that’s needed is a simple upgrade to get from here to
there, easy peasy!

Not so fast?! Well, OK, there are some “minor details” to sort in the process of getting there: Where is the
Goldilocks zone – not too little but not too much either – for the RF amps gain and power? There are
many items to consider.

With a comprehensive network model of a node + 6 cascade, should the “unity gain” concept be extended
all the way up to 1.8 GHz? Once modeled, what would an optimal power spectral density (PSD)
distribution across the forward spectrum look like? What happens as the distance between the amps
increases? Does the tap position on a link make a difference? What is the impact of the drop cables?
Should we continue to use end-of-line (EoL) throughput as our capacity benchmark?

There are many, many questions and this paper will start to look at some of the considerations with
rolling out 1794 MHz HFC plants. All of this is distilled into conclusions and best practices guidance – to
help make 10G and the DOCSIS 4.0 networks a reality for the time to come!

2. Unity Gain Overview

Figure 1 is a big picture view of a Hybrid Fiber Coaxial (HFC) network. It shows a “40,000-foot view” of
an HFC plant. While the plant and its architecture kept evolving – for example, a head end to node fiber
link took the place of a much longer cascade of trunk amps – it is also amazing how many things remain
the same: fiber nodes feeding RF amplifiers, amplifiers feeding taps, and taps, via “drops” delivering
signals to ever-important subscribers.

Figure 1 – A big picture view of a Hybrid Fiber Coaxial (HFC) network

Another HFC aspect that has remained the same is that of “unity gain” – a concept of setting every
amplifier output signal to the same shape and magnitude, no matter if the amp is first or last in the

© 2020 SCTE•ISBE and NCTA. All rights reserved. 4

cascade nor how “lossy” the span in front of the said amplifier was. This consistency of output is shown
via blue “amplifier output” lines in Figure 2 for the downstream direction.

Figure 2 – Illustration of the "Unity Gain" concept, applied to an RF amplifier cascade

In this example of a three-amplifier stage cascade, each output stage (in blue) is the same. However, the
coaxial cable and tap loss preceding each of the amplifiers is of a different value, as illustrated via green
“loss over frequency” lines. These various length cable spans, driven by the same previous-amplifier
output, will produce various inputs, shown via red “amplifier input” lines. It is the gain and slope of every
amplifier that is tuned, traditionally via selection of proper attenuator, cable simulator, and/or cable
equalizer, to make every amplifier output the same. One goal of the tuning is to set each amplifier’s input
stage signal as flat as possible to minimize noise figure contribution, the other is to up-tilt the final output,
to minimize non-linear distortion effects. Nevertheless, the noise effects and the non-linear distortions
still add up as the cascade length increases, despite keeping outputs of each amplifier the same.

Another rationale for unity gain is to provide a common input power and modulation order (e.g. 256-
QAM) for every cable consumer premises equipment (CPE) including legacy video set top boxes (STB)
as well as DOCSIS 2.0 &/or 3.0 cable modems (CM). This holds no matter what channel frequency is
being used or whether the CPE device is next to the node or at the end of line (EoL).

The span length between the amps, and how much of the corresponding amplifier gain is needed to
achieve “unity gain” has been the cable engineers’ focus for many years. Many HFC plants were
originally designed with amp spacing to accommodate lower frequencies, e.g. 450/550/750 MHz. Over
time, these were then “pushed” to higher frequencies, e.g. 870/1002/1218 MHz. Due to much higher cable
losses at higher frequencies, this resulted in amplifiers needing significantly higher output gains to
maintain the unity gain in the system. If the cable losses were more than the increased amplifier output
gains, then the HFC plant needed to be “re-spaced” where amplifier locations were moved, something
that is highly undesirable. This issue now gets exaggerated with pushing the HFC to 1.8 GHz.

As a point of reference, consider 750 MHz HFC plant that many “first world” countries built quite a few
years ago. At 750 MHz, these spans amount to ~20 dB attenuation following trunk amps, ~40 dB
attenuation following multi-port bridgers and ~30 dB attenuation following single output line extenders.

3. Network Capacity Planning for Cable 10G

The network capacity analysis in [ULM_2019] provides an insight into the capacity requirements needed
to support the cable 10G initiative. It first looks at the traffic engineering needed for a common 10G
network using both PON and cable systems. Then a closer look is taken at the spectrum planning for an
HFC system.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 5

The CommScope (formerly ARRIS) team has been providing industry leading research in traffic
engineering for many years which was recently highlighted in [ULM_2019]. Some additional references
of note include [CLO_2014], [EMM_2014], [ULM_2014], [CLO_2016], [ULM_2016], [ULM_2017] and
[CLO_2017].

3.1. The “Basic” Traffic Engineering Formula

Previously, [CLO_2014] introduced traffic engineering and quality of experience (QoE) for broadband
networks. From there, the paper went on to develop a relatively simple traffic engineering formula for
service groups that is easy to understand and useful for demonstrating basic network capacity
components.

The “Basic” formula shown below is a simple two-term equation. The first term (Nsub*Tavg) allocates
bandwidth capacity to ensure that the aggregate average bandwidth generated by the Nsub subscribers can
be adequately carried by the service group’s bandwidth capacity. The first term is viewed as the “DC
component” of traffic that tends to exist as a continuous flow of traffic during the peak busy period.

The “2014” Traffic Engineering Formula (Based on Tmax_max):

C ≥ (Nsub*Tavg) + (K*Tmax_max) (1)

where:
C is the required bandwidth capacity for the service group
Nsub is the total number of subscribers within the service group
Tavg is the average bandwidth consumed by a subscriber during the busy hour
K is the QoE constant (larger values of K yield higher QoE levels)…
where 0 ≤ K ≤ infinity, but typically 1.0 ≤ K ≤ 1.2
Tmax_max is the highest Service Tier (i.e. Tmax) offered by the MSO

There are obviously fluctuations that will occur (i.e. the “AC component” of traffic) which can force the
instantaneous traffic levels to both fall below and rise above the DC traffic level. The second term
(K*Tmax_max) is added to increase the probability that all subscribers, including those with the highest
service tiers (i.e. Tmax values), will experience good QoE levels for most of the fluctuations that go
above the DC traffic level.

The second term in the formula (K*Tmax_max) has an adjustable parameter defined by the K value. This
parameter allows the MSO to increase the K value and add bandwidth capacity headroom that helps
provide better QoE to their subscribers within a service group. In addition, the entire second term is
scaled to be proportional to the Tmax_max value, which is the maximum Tmax value that is being offered
to subscribers.

In previous papers [CLOONAN_2013, EMM_2014], found that a K value of ~1.0 would yield acceptable
and adequate QoE results. [CLOONAN_2014] goes on to provide simulation results that showed a value
between K=1.0 and 1.2 would provide good QoE results for a service group of 250 subscribers. Larger
service groups (SGs) would need even larger values of K while very small SGs might use a K value near
or less than 1.0.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 6

 3.2. Broadband Subscriber Traffic Consumption
CommScope/ARRIS has been monitoring subscriber usage for over a decade now from the same group of
MSOs. The data from this set has been compared and aligns closely to many other MSOs globally.

Figure 3 shows the average subscriber downstream consumption, DS Tavg, during peak busy hours for
several MSOs over a ten-year period. At the start of 2020, DS Tavg had surpassed the 2 Mbps barrier.

DS Avg BW per Subscriber

(2020 Avg = 2.36 Mbps, 3 yr Avg CAGR = 30.2%)
3500
3000
DS BW per Sub (kbps)

2500
2000
1500
2360
1000 1966
1621
500 823
1070
89 138 233 308 395 519
0
2010 2012 2014 Year 2016 2018 2020
DS Average MSO A MSO B MSO C MSO D

Figure 3 – Downstream Average Bandwidth per Subscriber through Jan ‘20

US Avg BW per Subscriber

(2020 Avg = 164 kbps, 3 yr Avg CAGR = 21.2%)
300
US Avg BW per Sub (kbps)

250

200

150

100
164
136 140
50 76 92
44 48 59 60
30 38
0
2010 2012 2014 Year 2016 2018 2020
Average MSO A MSO B MSO C MSO D

Figure 4 – Upstream Average Bandwidth per Subscriber through Jan ‘20

It turns out that the Tavg growth rate was higher at the start of this decade and has tailed off a bit in recent
years. Over the last 3-4 years, this group of MSOs had an average downstream traffic growth that had
been around 30%. Interestingly, the upstream traffic is growing at a significantly slower rate than the

© 2020 SCTE•ISBE and NCTA. All rights reserved. 7

downstream as shown in Figure 4. During the same ten-year period, the upstream Tavg generally grew at
less than 20% compound annual growth rate (CAGR).

This MSO data provides a good indication of Tavg, at least before the Coronavirus bandwidth (BW)
surge hit. Note that figures 3 and 4 are very generalized results that are averaged across millions of
subscribers.

Over recent years, there has been a slowing in the downstream usage growth rate (i.e. Tavg) compared to
the service tier growth rate (i.e. Tmax). This has several consequences including that the networks
become more “bursty”. It also means that the overall utilization of the network is lower too. In this
respect, it is important to try and maximize subscribers per service group (SG) in order to take advantage
of statistical multiplexing and get better economics.

3.3. Network Capacity Modeling for Cable 10G Downstream

[ULM_2019] used the CommScope network capacity modeling tools to see how a 1218/204 MHz HFC
plant could support the cable 10G downstream requirements. This analysis will first be reviewed and then
look at what changes for the extended spectrum DOCSIS 4.0 (ESD) scenarios.

The network capacity results from [ULM_2019] show the potential capabilities for a 1218/204 MHz HFC
plant. It begins with a 512-home passed (HP) service group with 256 subs (i.e. 50% penetration). The
max downstream (DS) service tier starts at 1 Gbps and grows by 1 Gbps per year from 2022 until it
finally reaches 8 Gbps DS SLA in the year 2028. For this case study, it is assumed the Tavg growth rate
continues its gradual decline over the next decade. This will leave Tavg at ~15 Mbps by the end of the
decade. If the Tavg growth rate does not decline, then these dates might get pulled in by 2-3 years.

Downstream DOCSIS Spectrum Required and Subs per SG

1056

960 250

864
Subs per SG
DOCSIS Spectrum (MHz)

768 200

672

576 150

480

384 100 3.0 SC-

QAM
288
D3.1
192 50 OFDM

96 HSD Subs
per SG
0 0
2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030

Figure 5 – 1218/204 MHz System – Subs per SG, DOCSIS Spectrum Needs

© 2020 SCTE•ISBE and NCTA. All rights reserved. 8

 Downstream DOCSIS Capacity & Breakdown

10000
IP Video - D3.1
9000
Tavg - D3.1
8000
D3.1 QoE Delta
7000
DS Tmax QoE -
6000
Mbps

3.0 Only
5000 Tavg - 3.0 Only

4000 IP Video - 3.0

3000 D3.1 + 3.0

Capacity
2000
3.0 DS Capacity
1000

0
2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030

Figure 6 – 1218/204 MHz System – DOCSIS Usage: Tmax, Tavg, IP Video

This 1218/204 MHz scenario shows that by 2027, all SGs need to be at 128 subs or fewer. The max subs
per SG is shown in Figure 5. The figure also breaks out the downstream spectrum amount needed for both
DOCSIS 3.0 SC-QAM and DOCSIS 3.1 OFDM. Note that by 2028 in this case study, 100% of DOCSIS
cable modems have been converted to DOCSIS 3.1 enabling OFDM channels across the entire spectrum
to maximize capacity.

The DOCSIS capacity usage is broken out in Figure 6. It shows the amount of capacity needed for both
DOCSIS 3.0 and DOCSIS 3.1. As can be seen, the Tmax component dominates over time. The upper red
line shows the combined 3.0 + 3.1 total capacity for the system.

Perhaps a key point of this case study was that a reasonably clean 512 HP node can be upgraded to
1218/204 MHz and support a service tier of 8 Gbps x 1.5 Gbps for the next decade. The only change
needed will be a SG segmentation (i.e. upgrade node from 1x1 to 2x2) somewhere in the middle of the
decade. There is no pressing near term need to push the HFC to very small (but inefficient!) SG sizes
found in N+0 systems.

3.4. Network Capacity Planning for DOCSIS 4.0 ESD

The cable 10G initiative strives for a more symmetric network with multi-gigabit upstream service tiers.
DOCSIS 4.0 ESD has evolved as a way to offer even higher upstream tiers while remaining in an N+X
environment. Thus, ESD also is trying to avoid the costs of an N+0 upgrade such as that needed with
DOCSIS 4.0 FDX.

3.4.1. Cable 10G Service Tiers

As shown in [ULM_2019], the subscriber is actually getting an 8 Gbps SLA in a 10G world after
accounting for all the different overheads (e.g. PHY, MAC, IP layers). Table 1 summarizes the various

© 2020 SCTE•ISBE and NCTA. All rights reserved. 9

options from that study and their respective downstream (DS) and upstream (US) SLAs that service
providers can consider offering. Because capacity in an HFC system can vary quite a bit based on many
variables, the offered SLAs are actually a range of values.

Table 1 – Summary of SLA Options for 10G PON & 1218 MHz Plants

DS SLA US SLA
10G PON Options (Gbps) (Gbps)
10G/1G EPON 8 0.8
10G/10G EPON 8 8
XG-PON 8 2
XGS-PON,
8 8
NG-PON2 (single wavelength)
DS SLA US SLA
10G HFC Options (Gbps) (Gbps)
1218/85 MHz 8 – 10 0.4 – 0.5
1218/204 MHz 6–8 1.0 – 1.5
1218/300 MHz 5–7 2.0 – 2.5
1218/396 MHz 4–6 2.5 – 3.0
1218/85 MHz +
8 – 10 5–6
108-684 MHz FDX/Soft-FDX

Perhaps the key motivating driver for DOCSIS 4.0 FDX and ESD is the ability to offer more symmetric
multi-gigabit upstream tiers. The minimum goal for ESD would be to at least match the DS tiers for a
1218/204 MHz plant while offering substantially higher upstream tiers. Ideally, ESD would be able to
match the DS tiers capabilities of 10G PON and 4.0 FDX systems.

The table above shows upstream splits up to 396 MHz. The DOCSIS 4.0 working group has also added a
492 MHz split and 684 MHz split as additional options.

3.4.2. DOCSIS 4.0 ESD – Tavg and Nsub targets

The [ULM_2019] study assumed that the Tavg growth rates continue to decline and estimated that Tavg
would reach 15 Mbps per sub by 2030. Given the additional investments needed for ESD, a Tavg of 20-
40 Mbps per sub will be considered to make sure there is additional lifetime for this investment.

Since ESD is targeted at N+X environments, our analysis will assume at least 125 subscribers per SG,
which was the ending point in the 2019 study.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 10

 3.4.3. DOCSIS 4.0 ESD – Network Capacity Requirements
The 2019 study showed that ~9 Gbps of DS capacity is required for 125 subs @ Tavg=15 Mbps. If Tavg
is bumped up to 25 Mbps, then total system capacity pushes above 10 Gbps. Pushing Tavg up to 40 Mbps
increases the required system capacity to around 12 Gbps. These are key targets that will be used in our
following analysis.

4. Simulation Model Overview

Our network performance estimates were done using a MATLAB simulation model. The model
simulated the effects of attenuation in the hardline coaxial cable, taps, drop cable and home entry port,
and amplification in the intermediate amplifiers, to calculate the received power levels at modems at
different locations. The effects of noise and distortion accumulation in the cascade were also simulated to
calculate the effective SNR at the modem. The results were compared with DOCSIS 4.0 received power
and SNR thresholds to estimate the available modulation and bit-loading at each frequency, yielding the
total gross throughput. Physical layer overheads such as cyclic prefix, FEC (LDPC), guard bands, pilot
and PLC tones were subtracted from the integrated bit-loading curve (gross throughput) to calculate the
net system throughput. The model used 78% efficiency for the PHY layer.

4.1. Cable Model – PSD and TCP

The DS power spectral density (PSD) at the node output was represented as a tilt/drop/tilt (a.k.a. lightning
bolt) in order to achieve a sensible total composite power (TCP) (nominally 70 dBmV based on 4.0
working group discussions, but this parameter could be varied if needed) while providing adequate levels
for legacy SC-QAM services at lower frequencies (up to 860 MHz). The PSD is shown in Figure 7.

Cable attenuation, for both hardline and drop cables, was calculated as a function of frequency and cable
segment length from tabulated data from cable manufacturers. A large database of tap response curves
from several tap families, taken from a combination of manufacturers’ data and our own measurements,
was available in the model. Note that no tap equalizers (or “cable simulators”) were present in the model;
all available RF power was used to optimize the bit-loading from each tap.

4.2. Amplifier Model

The amplifiers were modelled as non-ideal near-unity gain devices. The gain of each amplifier was tuned
to approximately match the attenuation of the preceding cable span, subject to a configurable maximum
gain. This approximate match used a linear-tilted gain over most of the spectrum, producing a small gain
deviation; over longer cascades, this produced a slightly U-shaped end-of-line PSD, in contrast to a flat
received PSD which would occur if perfect unity-gain amplification was possible.

The gain limit became significant for longer cable runs that is referred to as stretched plant. At higher
frequencies, the inter-amplifier span attenuation sometimes exceeded the assumed maximum gain,
resulting in progressively lower PSD at these frequencies after each stage in the cascade. Hence the
stretched plant did not quite achieve unity gain at 1800 MHz. Figure 8 shows an example of the inter-
amplifier attenuation and amplifier gain for a stretched plant.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 11

 Figure 7 – PSD Profile Showing Tilt/Drop/Tilt Characteristic

Figure 8 – Inter-amplifier Span Attenuation & Amp Gain (With 44 dB Gain limit) in a
Stretched Plant

© 2020 SCTE•ISBE and NCTA. All rights reserved. 12

 4.3. System Noise Performance
The noise performance of the system was modelled using an estimate of the node’s output SNR, and
noise figures for the amplifiers in the cascade, and for the cable modem. The various noise components
were modelled through the successive attenuation and amplification stages of the cascade to produce a
received noise estimate. Distortion was treated in a similar way; distortion at the node output and each
amplifier output was modelled as a third-order function of TCP, and coherence effects were also
calculated.

4.4. Amplifier Spacing – Typical and Stretch Networks

For our network simulations, it was important to model amplifier spacings that were representative of
real-world data based on customer’s amplifier spacing data. A cumulative distribution function (CDF) of
some sample amplifier spacing for single output line extenders (LE) is shown in Figure 9, while the CDF
for multi-port bridgers (MB) is shown in Figure 10. Line extenders often account for more than 60% the
total number of amplifiers. This data is courtesy of Cox Communications.

Most of the simulations used either “typical plant” or “stretch plant” parameters for plant length. “Typical
plant” parameters are chosen to include a majority of inter-amplifier span attenuations (for line-extender
(LE) spacings). A total attenuation of 30 dB at 1 GHz and 40 dB at 1794 MHz is used, which is an
attenuation value greater than that of ~65% of line-extender to line-extender spacings according to Figure
9. This attenuation is simulated with 6 segments of 175’ of P3 625 hard-line and 5 taps, a total of 1050’
between amplifiers. Note that 850’ of P3 500 would give similar results.

Unless otherwise specified, each simulation used the same parameters for the node to first amplifier span
and for the last amplifier to end tap as for the inter-amplifier spans.

Line Extender Spacing at 1.8 GHz*

7% 100%
90%
6%

C umulative Percentage
80%
Percent of Spacings

5% 70%
4% 60%
50%
3% 40%
2% 30%
20%
1%
10%
0% 0%
19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
dB Spacing
* Estimated - by adding 40% to 1 GHz coax loss and 20% to 1 GHz passive loss

Figure 9 – Amplifier Spacing – Single Output Line Extenders (LE)

© 2020 SCTE•ISBE and NCTA. All rights reserved. 13

 Multiport Bridger Amp Spacing at 1.8 GHz*
6% 100%
90%
5%

C umulative Percentage
80%
Percent of Spacings

4% 70%
60%
3% 50%
40%
2% 30%
1% 20%
10%
0% 0%
27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61
dB Spacing
* Estimated - by adding 40% to 1 GHz coax loss and 20% to 1 GHz passive loss

Figure 10 – Amplifier Spacing – Multi-Output Bridgers (MB)

“Stretch plant” parameters are chosen to exceed most inter-amplifier span attenuations; the corresponding
attenuations are 35 dB at 1 GHz and 47 dB at 1794 MHz. This is greater than ~97% of LE-to-LE
spacings, and two-thirds of bridger-to-amplifier spacings according to Figure 10. Overall, this would
cover more than 85% of the amplifier links. This attenuation is simulated with 6 segments of 215’ of P3
625 hard-line and 5 taps, a total of 1290’ between amplifiers. Note that 1050’ of P3 500 would give
similar results.

For both typical and stretch plants, the baseline case used 150’ of RG-6 drop cable along with a 3.5 dB
loss inside the home (e.g. splitter at point of entry or 20’ of RG-6 inside the home). Note that cable losses
from 150’ of RG-6 drop cable would be roughly equivalent to 210’ of RG-11 drop cable.

5. Conventional Wisdom Using Unity Gain Approach

As discussed previously, unity gain helped maintain a constant QAM-modulation for legacy STB and
DOCSIS 2.0/3.0 cable modems. System performance was defined by the lowest common denominator.
The end of line (EoL) provided a key performance monitoring spot to determine whether every CPE
device could successfully receive the 256-QAM downstream channel. But future systems will be
migrating to DOCSIS 3.1 and 4.0 cable modems. Our network simulations calculate the maximum
downstream network capacity achievable using 3.1/4.0 OFDM channels with variable-bit loading per sub-
carrier with up to 4096-QAM modulation.

For our network simulations, the EoL performance is first checked for both 1218 and 1794 MHz systems
to see how well unity gain performed. Figure 11 looks at the network capacity for a 108-1218 MHz
downstream for various amplifier cascade lengths, from Node+0 up to Node+6. This was done for both a
typical plant and a stretch plant conditions.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 14

 End-of-Line Throughput - 108-1218 MHz, 150' Drop cables
12

11
Net Throughput [Gbps]

10

7
Node+0 Node+1 Node+2 Node+3 Node+4 Node+5 Node+6
Cascade

108-1218 Typical EoL 108-1218 Stretch EoL

Figure 11 – End-of-Line Throughput - 108-1218 MHz, 150' Drop cables

Figure 12 – Stretch 108-1218 MHz Plant Bit-loading – Tap 5 for N+2 and N+5

The 1218 MHz system maintains unity gain for both the typical and stretch plants. The network capacity
degrades slightly as the amplifier cascades increases due to the accumulated noise and distortion from
each additional amp stage. The additional cable loss in the stretch plant degrades network capacity
slightly from the typical plant, but each amplifier stage is still able to maintain its unity gain.

Figure 12 shows the bit-loading for the 5th tap in a N+2 and N+5 stretch 108-1218 MHz system. As can
be seen, the bit-loading remains fairly constant over the entire spectrum gain. This is what one might
expect with unity gain.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 15

Figure 13 now shows the EoL throughput for a 492-1794 MHz downstream system with cascade lengths
varying from Node+0 to Node+6. This has 192 MHz more total downstream spectrum than the 108-1218
MHz system, but it is spectrum at higher frequencies with higher losses. Note that the downstream
network capacity of the 492-1794 MHz typical plant in Figure 13 is very close to the 108-1218 MHz
typical plant in Figure 12. However, the EoL throughput for the 1.8 GHz stretch plant in Figure 13 is
substantial lower than the typical plant by around 20% or almost 2 Gbps. The stretch plant is also below
the minimum 9 Gbps capacity target that was discussed in section 3.

Figure 14 shows the bit-loading for the 5th tap in a N+2 and N+5 typical 492-1794 MHz system. Since
unity gain is maintained, there is no drop-off in bit-loading at higher frequencies. In fact, there is a slight
increase in the bit-loading at the higher frequencies due to some non-linear effects in the amplifier.

End-of-Line Throughput - 492-1794MHz, 150' Drop cables

12

11
Net Throughput [Gbps]

10

7
Node+0 Node+1 Node+2 Node+3 Node+4 Node+5 Node+6
Cascade

Typical Plant Stretch Plant

Figure 13 – End-of-Line Throughput - 492-1794 MHz, 150' Drop cables

Figure 14 – Typical 492-1794 MHz Plant Bit-loading – Tap 5 for N+2 and N+5

© 2020 SCTE•ISBE and NCTA. All rights reserved. 16

 Figure 15 – Stretch 492-1794 MHz Plant Bit-loading – Tap 5 for N+2 and N+5

The stretch 492-1794 MHz system does not maintain unity gain. Figure 8 showed that the amplifier gain
reaches its maximum output around 1600 MHz. Figure 15 shows the bit-loading for the 5th tap in a N+2
and N+5 stretch 492-1794 MHz system. It clearly shows how the bit-loading starts to drop-off above
1600 MHz. Note that the loss is relatively minimal after the 2nd amplifier output, only 1-2 orders of
modulation. However, by the output of the 5th amplifier, the 5th tap is seeing a dramatic drop in bit-
loading, losing roughly half the capacity in the 1602-1794 MHz OFDM channel.

Based on our earlier network capacity planning targets of needing 9-12 Gbps, the stretch 492-1794 MHz
plant would not have acceptable system capacity using EoL throughput measure. Our conventional
wisdom would dictate that these plants would need to have the amplifiers re-spaced or a mid-span
amplifier added in order to maintain unity gain.

6. Game Changer – DOCSIS Compensates When Unity Gain Lags

When the DOCSIS 3.1 (D3.1) specification was written, the authors had a vision of a new DOCSIS PHY
layer that squeezed every last once of capacity out of the cable plant. DOCSIS 4.0 inherits all these
capabilities. Some of the most important D3.1 capabilities were wide OFDM channels with variable bit-
loading along with the support of multiple profiles.

Previously, all modems would receive the identical downstream data stream. Therefore, the entire
network had to operate for the lowest common denominator. Hence, the EoL capacity was a good metric
for determining the plants capabilities. Each OFDM channel has a group of D3.1 profiles. Within that
channel, modems are put into profile groups with similar modems and receive data at an optimal data rate.
For example, modems in one profile could be receiving at 4096-QAM modulation while other modems in
a different profile are receiving at 256-QAM modulation. This allows the DOCSIS system to optimize
system capacity. This means that total system capacity depends on the capacity seen by every home on
every tap in the system.

Figure 16 shows a 108-1218 MHz downstream example in a N+6 plant for both typical and stretch
amplifier spacing. Network capacity is shown at every tap in the system. Modems further out on the
cascade are seeing slightly lower performance and would be operating with D3.1 profiles that are using
QAM modulations one or two steps below the best profile. In this example, the best profile would have
20% more capacity than the lowest profile.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 17

 Net Throughput vs Tap Position in 108-1218 MHz Node+6 Plant
12

11
Net Throughput [Gbps]

10

7
0_1 0_2 0_3 0_4 0_5 1_1 1_2 1_3 1_4 1_5 2_1 2_2 2_3 2_4 2_5 3_1 3_2 3_3 3_4 3_5 4_1 4_2 4_3 4_4 4_5 5_1 5_2 5_3 5_4 5_5 6_1 6_2 6_3 6_4 6_5
Tap Number

108-1218 MHz (stretch) 108-1218 MHz (typical)

Figure 16 – 108-1218 MHz Net Throughput vs. Tap Position in N+6 Plant, 150' Drop
cables

492-1794 MHz Throughput vs Tap Position in N+6 Typical Plant, 150' Drop cables
12

10
EoL EoL EoL EoL
N+0 N+1 N+2 N+3 EoL EoL EoL
8 N+4 N+5 N+6
Throughput [Gbps]

0
0_1 0_2 0_3 0_4 0_5 1_1 1_2 1_3 1_4 1_5 2_1 2_2 2_3 2_4 2_5 3_1 3_2 3_3 3_4 3_5 4_1 4_2 4_3 4_4 4_5 5_1 5_2 5_3 5_4 5_5 6_1 6_2 6_3 6_4 6_5 6_6

Tap Number

Typical Plant

Figure 17 – 492-1794 MHz Throughput vs. Tap Position, N+6 Typical Plant, 150' Drops

© 2020 SCTE•ISBE and NCTA. All rights reserved. 18

For 1.8 GHz plants, the cable losses at the higher frequencies will magnify the differences between the
best profiles and the lowest profiles. Figure 17 helps give a glimpse into the wider discrepancy seen by
modems across an entire typical plant. Capacity changes based on two factors: # of amplifiers passed; and
# of taps from the amplifier output. Figure 18 shows the stretch plant that has even more of a swing. In
this example, the typical plant has a 32% swing from the lowest to highest capacities while the stretch
plant sees a 60% swing from low to high! It goes from ~7.5 Gbps at tap 5 on the 6th amplifier up to almost
12 Gbps at the 1st tap from the node.

Remember, the EoL capacities discussed earlier basically aligns with the 5th tap value after the last
amplifier, so it is basically following the troughs on this chart (see EoL for each N+X in Figures 17 &
18). These wide variations are just waiting to be optimized by DOCSIS. Many of the modems will have
capacities significantly above the EoL capacities.

The profiles are set up separately for each DOCSIS OFDM channel. The OFDM channels are up to 192
MHz wide. So, the 492-1794 MHz downstream will have up to 7 OFDM channels with each channel
getting its own optimized profiles.

In addition to optimizing each channel, the CMTS scheduler does load balancing between the channels.
And an intelligent scheduler can factor the profiles for each modem to determine which channels it should
use to optimize total network capacity.

So, if EoL is not a good measure of the network capacity, what should be used? It turns out that this is an
extremely complex answer based on many different variables. The following sections will explore a
number of these variables and make some estimates of network capacity for a ‘reasonable’ system.

492-1794 MHz Throughput vs Tap Position in N+6 Stretch Plant, 150' Drop cables
12

10

8 EoL EoL EoL

N+1 EoL
Throughput [Gbps]

N+0 N+2 EoL

N+3 EoL EoL
N+4 N+5
6 N+6

0
0_1 0_2 0_3 0_4 0_5 1_1 1_2 1_3 1_4 1_5 2_1 2_2 2_3 2_4 2_5 3_1 3_2 3_3 3_4 3_5 4_1 4_2 4_3 4_4 4_5 5_1 5_2 5_3 5_4 5_5 6_1 6_2 6_3 6_4 6_5 6_6

Tap Number

Stretch Plant

Figure 18 – 492-1794 MHz Throughput vs. Tap Position, N+6 Stretch Plant, 150' Drops

© 2020 SCTE•ISBE and NCTA. All rights reserved. 19

 6.1. Estimating HP/Subscriber Distribution across N+X HFC Plant
As can be seen in Figures 17 & 18 above, the available capacity to any given modem is a function of
where it is located on the plant relative to the fiber node. The HFC plant is a tree and branch topology that
fans out for each additional level in the cascade. If one blindly assumed that each amplifier has 2 outputs
feeding the next level of amplifiers, then a N+6 plant would have 64 amplifiers at the last stage in the
cascade and a total of 126 amplifiers per node. If the outputs per amplifier is increased to 3 or 4, then it
grows exponentially to where there could be 1000’s of amplifiers from a given node.

But this is not reality. To understand what is out in the real world, the authors spoke with our customers
and tapped into the knowledge of the in-house CommScope HFC design team. This in-house design team
is perhaps the most experienced HFC design team anywhere, with a legacy going back 30+ years at some
of our previous incarnations: ARRIS, C-Cor, Motorola, General Instrument, Philips.

The first thing we learned is that there is really no “typical” system. Real world HFC plants vary all over
the map. So, we set out to define a “reasonable” scenario to get some baseline capacity estimates. From
there, certain variables can then be changed to understand their impact on network capacity.

In general, amplifier spacing on average is relatively constant at 4-5 amplifiers per mile – especially if the
networks considered include a mix of low density rural, medium density suburban and high-density urban
areas. The homes passed (HP) density impacted how many homes might be off each tap (e.g. 1-2 HP per
rural tap, 4-8 HP per urban tap). So, for our analysis, it is assumed that HP would be distributed evenly
across all taps in the system.

As a starting point, a suburban build with 80-110 homes passed per mile (HP/mile) was chosen. From
there, plants of various cascade lengths (i.e. N+0 to N+6) were analyzed. Table 2 shows an estimate of RF
amps and homes passed based on input from our expert in-house HFC design team. This table shows that
the number of RF amplifiers per node grows almost linearly with the cascade depth. It is NOT
exponential! For each level in the cascade, another six to eight amplifiers are typically added.

Why is this??? It turns out there are several factors. First, the majority of amplifiers are single output line
extenders (LE). These are amplifiers just cascading down a street without any branches. Next, multiple
output bridgers (MB) would often have some outputs feeding side streets without connecting to another
amplifier (i.e. a dead end). Finally, not all branches go out the entire cascade depth. In a N+6 plant for
instance, many branches might terminate after 3, 4 or 5 amplifiers and not reach the sixth cascade level.

The homes passed per node also tended to follow the amplifier count with typically 16-20 HP per amp.

This knowledge then let us model how the homes passed (and hence subscribers) are spread across the
HFC plant. After the first amplifier, a linear increase in HP per cascade level is used.

The next task was to understand how many taps are typically following each amplifier. As noted above,
the amplifier spacing is fairly constant, and hence the number of taps were too. But there is some
variation. The HFC design team might need to shorten or extend a leg based on the real-world geography.
The vast majority of amplifiers have either 4 or 5 taps at its outputs. The number of fewer taps (e.g. 3) or
more taps (e.g. 6) was trivial for this analysis. It turns out that percentage of amplifiers with 4 taps is
roughly twice that of the amplifiers with 5 taps.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 20

 Table 2 – Example RF Amps and Homes Passed per Node for N+X

HFC Cascade # of RF Amps # of Homes Passed

Lengths per Node per Node
Node + 0 0 40 - 60
Node + 1 4–6 80 - 150
Node + 2 10 – 12 150 - 240
Node + 3 15 – 20 240 - 360
Node + 4 22 – 28 360 - 480
Node + 5 28 – 35 480 - 600
Node + 6 35 - 45 600 - 720

6.2. DOCSIS Optimized Network Capacity Analysis

The first order of business was to look at the 108-1218 MHz plant shown in Figure 16 and see how much
additional DOCSIS capacity might be available compared to the EoL measurements. Figure 19 shows the
DOCSIS optimized weighted average capacity compared to the EoL capacity for both the typical and the
stretch 1218 MHz plants.

End-of-Line vs. Weighted Avg Throughput - 108-1218 MHz, 150' Drop cables
12

11
Net Throughput [Gbps]

10

7
Node+0 Node+1 Node+2 Node+3 Node+4 Node+5 Node+6
Cascade

108-1218 Typical EoL 108-1218 Typical Avg 108-1218 Stretch EoL 108-1218 Stretch Avg

Figure 19 – End-of-Line vs. Weighted Avg Throughput - 108-1218 MHz, 150' Drop cables

© 2020 SCTE•ISBE and NCTA. All rights reserved. 21

 End-of-Line vs. Weighted Avg Throughput - 492-1794 MHz, 150' Drop cables
12

11
15%
35%
Net Throughput [Gbps]

10

7
Node+0 Node+1 Node+2 Node+3 Node+4 Node+5 Node+6
Cascade

492-1794 Typical EoL 492-1794 Typical Avg 492-1794 Stretch EoL 492-1794 Stretch Avg

Figure 20 – End-of-Line vs. Weighted Avg Throughput - 492-1794 MHz, 150' Drop cables

Both typical and stretch plants saw very little gain for shorter cascades while N+6 plant saw up to 10%
gains over the EoL calculations. So EoL appears to be reasonably accurate for the 1218 MHz plant.

The next scenario looked at the 492-1794 MHz plant previously shown in Figure 17 & 18 to see how
much additional DOCSIS capacity might be available compared to the EoL measurements. Figure 20
shows the DOCSIS optimized weighted average capacity compared to the EoL capacity for both the
typical and the stretch plants.

The difference between the DOCSIS optimized weighted average capacity and EoL capacity is
significant. For the typical plant, the gains start around 10% for N+0 and increase to 15% for N+6. The
stretch plant sees even larger gains, from 25% gains for N+0 plant to 35% for N+6 plant. The EoL
measure is grossly underestimating the capacity of the stretch plant. The EoL measure showed that typical
plants had 20% to 25% more capacity than the stretch plant. The DOCSIS optimized capacity results
above shows that the gap between typical and stretch plant is significantly less, in the 5% range!

Perhaps one of the most significant conclusions from this study is that the stretch plant that was
considered inadequate, is actually capable of supporting the network capacity requirements and does NOT
need any extra help (e.g. re-spacing amps or adding mid-span amplifiers).

6.3. Impact of shorter drop cables (66’/100’/150’)

Drop cables are less costly than trunk lines, but also do not have as good RF characteristics. This becomes
a big deal as the HFC plant pushes to 1.8 GHz. Per our in-house HFC design expert, most drop cables out
there today are RG-6 cables. Higher performance RG-11 drop cables are typically used for extremely long
runs, e.g. 200’ or longer.

Our expert felt that our 150’ drop cable assumption in the previous results was on the high side. Figure 21
and 22 show the impact on total network capacity by varying the drop cable length from 66’ to 100’ to

© 2020 SCTE•ISBE and NCTA. All rights reserved. 22

150’. This was done for the typical plant in Figure 21 and the stretch plant in Figure 22, using RG-6 cable
for all cases.

Net Throughput vs Tap Position in N+6 Typical Plant, 66/100/150 ft drop cable

12

11
Net Throughput [Gbps]

10

7
0_1
0_2
0_3
0_4
0_5
1_1
1_2
1_3
1_4
1_5
2_1
2_2
2_3
2_4
2_5
3_1
3_2
3_3
3_4
3_5
4_1
4_2
4_3
4_4
4_5
5_1
5_2
5_3
5_4
5_5
6_1
6_2
6_3
6_4
6_5
6_6
Tap Number

Typical 66' Typical 100' Typical 150'

Figure 21 – Net Throughput vs Tap Position in N+6 Typical Plant, 66/100/150 ft drop cable

Net Throughput vs Tap Position in N+6 Stretch Plant, 66/100/150 ft drop cable

12

11
Net Throughput [Gbps]

10

7
0_1
0_2
0_3
0_4
0_5
1_1
1_2
1_3
1_4
1_5
2_1
2_2
2_3
2_4
2_5
3_1
3_2
3_3
3_4
3_5
4_1
4_2
4_3
4_4
4_5
5_1
5_2
5_3
5_4
5_5
6_1
6_2
6_3
6_4
6_5
6_6

Tap Number

Stretch 66' Stretch 100' Stretch 150'

Figure 22 – Net Throughput vs Tap Position in N+6 Stretch Plant, 66/100/150 ft drop cable

© 2020 SCTE•ISBE and NCTA. All rights reserved. 23

In general, the short drop cable makes a significant difference, especially for the fourth and fifth taps. For
the typical plant, there is less than a 3% difference on the first two taps after the amplifier for 100’
compared to 150’. The 4th tap has gains in the 7-10% range while the 5th tap sees 7-15% improvements.

The stretch plant shows even wider spreads between 100’ and 150’. The gains on the first two taps can be
up to 4%. The 4th tap sees improvements up to the 8-12% range and the 5th tap sees 10-16% gains with the
shorter 100’ drops.

Note that replacing a 150’ RG-6 drop cable with RG-11 will result in improved performance that
is very close to the 100’ RG-6 drop cable.

Migrating from 100’ to even shorter 66’drop cables did not have as big effect, especially on the typical
plant. The 66’ drop cable did show some gains over the 100’ drop on the 4th and 5th taps of the stretch
plant. The largest gains being 7-10% after the first two amplifier and then shrinking gains with more
amplifiers in the cascade.

6.4. Impact of Super-stretched Links (56dB)

As shown on Figure 10, a small percentage of multi-port bridger amplifiers have extremely large dB
spacings that are >55 dB. Figure 23 shows the impact of inserting one or two of these super-stretched
links with 56 dB spacing into a typical plant. For reference, 56 dB spacing between amplifiers could
represent any one of the following scenarios:

1. 1560 ft of 0.500 cable

2. 1920 ft of 0.625 cable
3. 2640 ft of 0.875 cable
4. Or 1560 ft of 0.625 plus 5 taps

In general, it would be expected that these super stretched links are express feeder runs over a long
distance without any taps. For this example, Figure 23 shows scenario 4 with 1560’ of 0.625 cable plus 5
1.8 GHz taps to demonstrate the impacts of the distance on these taps.

For the first case, a super-stretch 56 dB link is inserted following the node. As can be seen, the total
capacity seen at the output of the 1st amplifier drops by ~0.5 Gbps at the 1st tap and over 1 Gbps by the 5th
tap relative to the typical plant. After each additional amplifier in the cascade, the delta with the typical
plant shrinks as the noise introduced by each amplifier accumulates and starts to dominate.

For the second case, the super-stretch 56 dB link is inserted after the 3rd amplifier. The outputs after the
node, 1st and 2nd amplifiers are identical to the typical plant. The outputs after the 4th, 5th and 6th amplifiers
are very close to the first case in which the super-stretch link is at the node.

The final case shows a scenario where two super-stretch links are inserted, one after the node and one
after the 3rd amplifier output. Adding the second super-stretch link shows a noticeable degradation to the
total network capacity after the 4th, 5th and 6th amplifier outputs.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 24

 492-1794 MHz Typical Plant with 1 or 2 Super-stretch 56dB Links
12

Net Throughput (Gbps) 11

10

7
0_1
0_2
0_3
0_4
0_5
1_1
1_2
1_3
1_4
1_5
2_1
2_2
2_3
2_4
2_5
3_1
3_2
3_3
3_4
3_5
4_1
4_2
4_3
4_4
4_5
5_1
5_2
5_3
5_4
5_5
6_1
6_2
6_3
6_4
6_5
6_6
stretch@node Typical Plant

492-1794 MHz Typical Plant with 1 or 2 Super-stretch 56dB Links

12

11
Net Throughput (Gbps)

10

7
0_1
0_2
0_3
0_4
0_5
1_1
1_2
1_3
1_4
1_5
2_1
2_2
2_3
2_4
2_5
3_1
3_2
3_3
3_4
3_5
4_1
4_2
4_3
4_4
4_5
5_1
5_2
5_3
5_4
5_5
6_1
6_2
6_3
6_4
6_5
6_6

stretch@N+3 Typical Plant

492-1794 MHz Typical Plant with 1 or 2 Super-stretch 56dB Links

12

11
Net Throughput (Gbps)

10

7
0_1
0_2
0_3
0_4
0_5
1_1
1_2
1_3
1_4
1_5
2_1
2_2
2_3
2_4
2_5
3_1
3_2
3_3
3_4
3_5
4_1
4_2
4_3
4_4
4_5
5_1
5_2
5_3
5_4
5_5
6_1
6_2
6_3
6_4
6_5
6_6

stretch@node & N+3 Typical Plant

Figure 23 – 492-1794 Typical Plant with 1 or 2 Super-stretch 56 dB Links

© 2020 SCTE•ISBE and NCTA. All rights reserved. 25

 DS Weighted Avg Throughput - for Typical Plant vs. Super-stretch Plants, 150' Drop cables
12

11
Net Throughput [Gbps]

10

Typical Plant
stretch@N+3
stretch@node
9
stretch@node & N+3

7
Node+0 Node+1 Node+2 Node+3 Node+4 Node+5 Node+6
Cascade

Figure 24 –Weighted Average DS Capacity with 1 or 2 Super-stretch 56 dB Links

Figure 24 compares the DOCSIS optimized weighted average DS capacity for the three super-stretch
cases compared to the typical plant. With the CMTS scheduler effectively averaging capacity across all
the taps, this helps to minimize the impact of these super-stretched links. For Node+6 plant, adding one
super-stretched link drops average capacity by <5% while adding two super-stretched links only has
~10% impact.

6.5. Downstream ESD Capacity for various US/DS splits

The DOCSIS 4.0 specification allows for many different upstream (US) options. For a static ESD system,
the possible diplexer splits are:

• 85/108 MHz
• 204/258 MHz
• 300/372 MHz
• 396/492 MHz
• 492/606 MHz
• 684/834 MHz

The first number is the upper edge of the upstream spectrum. The second number is the starting edge of
the downstream. The guard band is the region between these two and is not usable. Notice that the guard
band grows larger as the US/DS split frequencies increase.

Our network capacity modeling research has shown time and again that downstream bandwidth capacity
is the limiting factor in almost every analysis. So, it is extremely important to try and maximize the
downstream bandwidth capacity.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 26

The DOCSIS 4.0 specification also supports a dynamic FDX option. For FDX, the shared band starts at
108 MHz and can go up to 204, 300, 396, 492 or 684 MHz. The FDX downstream goes up to 1218 MHz.
So downstream bursts can potentially leverage the full 108-1218 MHz.

Figure 25 shows the downstream capacity at each tap location for various static ESD US/DS splits for a
N+6 typical plant and 150’ drop cables. It also compares these to a 108-1218 MHz DS similar to that used
by FDX (although full-duplex operation may add other impairments which are not factored into this
example). Figure 26 then shows the DOCSIS optimized DS weighted average capacity for the various
static ESD splits and the dynamic FDX for various cascade lengths from N+0 to N+6.

Because the changes between the US/DS splits are in the lower frequencies (i.e. below 684 MHz), the
capacity differences remain constant. For each static ESD curve, there is a drop in capacity with
increasing taps. This is the same for all static ESD scenarios, so they all have the same shape. Note that
the 684/834 split has a noticeably larger gap than the other splits. This is a combination of two factors.
First, it has the largest guard band. Second, its downstream is missing a lot of the lower frequencies which
tend to have higher bit-loading as shown previously in Figures 14 and 15.

The 108-1218 MHz FDX DS spectrum is also shown for reference and appears to align closest with the
492/606 MHz static ESD DS capacity. The next section takes a closer look at that.

Net DS Throughput vs Tap Position in Node+6 Typical Plant for different US/DS splits
16

14

108-1218 MHz
12
108-1794 MHz

258-1794 MHz
10
Net Throughput [Gbps]

372-1794 MHz

492-1794 MHz

8 606-1794 MHz

834-1794 MHz

0
0_1
0_2
0_3
0_4
0_5
1_1
1_2
1_3
1_4
1_5
2_1
2_2
2_3
2_4
2_5
3_1
3_2
3_3
3_4
3_5
4_1
4_2
4_3
4_4
4_5
5_1
5_2
5_3
5_4
5_5
6_1
6_2
6_3
6_4
6_5

Tap Number

Figure 25 – Net Throughput vs. Tap Position in N+6 Typical Plant, different US/DS splits

Table 3 looks at some potential service tier combinations that could be considered for the various ESD
split options. These SLAs are based on the basic traffic engineering formula discussed earlier in section 3.
Remember that it is a function of the number of subscribers, Nsub, and the average peak period
consumption per sub, Tavg. Both may change over time. For this table, the base assumption is Nsub=200,
DS Tavg=20 Mbps, US Tavg=1.2 Mbps and K=1.0-1.2. Note that 100 subs @ DS Tavg=40 Mbps would
be roughly equivalent.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 27

By comparison, the SLA for 10G PON with 64 subs and same Tavg would be 6-7.5 Gbps for both the US
+ DS. The last row in Table 3 shows the theoretical best SLA’s on a 1794 MHz plant. This uses dynamic
soft-FDD where the CMTS scheduler switches between two diplexer settings (i.e. 85/108 MHz and
684/834 MHz) to time division multiplex the 108-684 MHz spectrum between US + DS.

DS Weighted Avg Throughput - for all 1794 MHz splits, Typical Plant, 150' Drop cables
16

14

12
Net Throughput [Gbps]

10 108-1218 MHz

834-1794 MHz
8
606-1794 MHz
492-1794 MHz

6 372-1794 MHz
258-1794 MHz
108-1794 MHz
4

0
Node+0 Node+1 Node+2 Node+3 Node+4 Node+5 Node+6
Cascade

Figure 26 – Weighted Average DS Capacity for different US/DS splits

Table 3 – Summary of SLA Options for 1794 MHz Plants

Top of US Start of DS DS SLA US SLA

10G HFC Options (MHz) (MHz) (Gbps) (Gbps)
1794/204 MHz 204 258 7–9 1 – 1.5
1794/300 MHz 300 372 6–8 1.5 – 2
1794/396 MHz 396 492 5–7 2.5 – 3
1794/492 MHz 492 606 4–6 3 – 3.75
1794/684 MHz 684 834 3–4 4–5
1794/85 MHz with
684 108 8 – 10 4–5
108-684 MHz dynamic Soft-FDD

© 2020 SCTE•ISBE and NCTA. All rights reserved. 28

 6.6. Comparing 108-1218MHz FDX vs. 606-1794MHz ESD DS Capacity
The DOCSIS optimized capacity for both the static 606-1794 MHz ESD and the 108-1218 MHz dynamic
FDX DS capacities are compared in Figure 27. This gives a better handle on how these two stacks up
against each other. Note – FDX in N+X plant would require FDX-capable amplifiers which are still under
investigation.

For typical plant conditions, the DOCSIS optimized weighted average capacity for the 606-1794 MHz
ESD is almost identical to the 108-1218 MHz FDX system.

On a stretch plant, the 606-1794 MHz ESD takes a hit in the higher frequencies, especially above 1.5
GHz. This reduces its DS capacity by about 5% compared to the 108-1218 MHz DS in a stretch plant.

Weighted Avg Throughput - 108-1218 MHz vs 606-1794 MHz, 150' Drop cables
12

11
Net Throughput [Gbps]

10

7
Node+0 Node+1 Node+2 Node+3 Node+4 Node+5 Node+6
Cascade

108-1218 Typical Avg 108-1218 Stretch Avg 606-1794 Stretch Avg

Figure 27 – Weighted Avg Throughput - 108-1218 MHz vs. 606-1794 MHz, 150' Drop cables

6.7. Improving on the DOCSIS Weighted Average Capacity

The DOCSIS weighted average capacity in the earlier analysis’s assumes that the subscriber’s data usage
is relatively evenly distributed across the HFC plant AND the subscriber’s data usage is evenly
distributed across all the available OFDM channels. However, each channel has its own unique set of
profiles that can vary significantly from channel to channel.

Figure 28 below looks at the bit loading for two extreme cases in the 492-1794 MHz plant. The first case
is a home on the 1st tap after the 1st amplifier for a typical plant amp spacing. The second case is a home
on the 5th tap after the 6th amplifier on a stretch plant.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 29

 Figure 28 – Example Bit-loading for Near and Far Homes

As can be seen, the modem at the far home has vastly lower bit-loading capacity in the upper frequencies,
and in particular, the top OFDM channels, i.e. 1602-1794 MHz. If the far modem has a 5 Gbps burst, the
CMTS could schedule it in the 492-1218 MHz OFDM channels. Note that the near modem still has ~5
Gbps available to it in the top 3 OFDM channels from 1218-1794 MHz. So, both modems could be
bursting to 5 Gbps simultaneously even though the EoL throughput for this plant is less than 7.5 Gbps.
And the total burst rate is higher than the average of the two bit-load maps put together.

OFDM Channel Capacity per Tap Position - Stretch Plant, 150' Drop
12

10

8 1602-1794
Channel Caapacities (Gbps)

1410-1602

6 1218-1410

1026-1218

4 834-1026

642-834

492-642
2

Tap Position

Figure 29 – Example Channel Capacities for various Tap Positions

© 2020 SCTE•ISBE and NCTA. All rights reserved. 30

Figure 29 shows how these bit-loadings map to the individual channel capacity for the 1st, 3rd and 5th tap
on every amplifier output stage. As can be seen in the above channel capacities, the top OFDM channel
for the 5th tap is practically unusable. For the other high frequency OFDM channels (i.e. 1026-1602 MHz)
the 1st and 3rd taps can be 30% to50% higher capacities than the 5th tap location. Overall, the 1st tap
position is getting reasonably good capacity across all the channels, even at the higher frequencies.

For over a decade, the CMTS schedulers have been load balancing across multiple DOCSIS 3.0 bonded
channels. And they have been doing a great job at it. In the above scenario, the CMTS scheduler can tend
to put data from the further taps in the lower frequency OFDM channels while the nearer taps utilize the
relatively empty upper frequency OFDM channels. This means the CMTS can achieve even higher
capacities than those shown with the weighted average capacity analysis above.

6.8. Handling the Corner Cases – Meeting Tmax Burst QoE

The DOCSIS weighted average capacity and the intelligent CMTS scheduler improvements addresses the
total system capacity available to the scheduler. However, as seen previously, the capacity seen by any
individual home can vary dramatically. The CommScope basic traffic engineering formula discussed in
Section 3 can help determine the capacity needed for each individual home so it can obtain its appropriate
Quality of Experience (QoE).

Consider a worst-case scenario where an operator wants to support a 7.5 Gbps DS service tier across their
entire footprint, including 10G PON FTTH and DOCSIS 4.0 ESD 492/1794 MHz HFC plant. As seen in
Figure 18, some home locations might only see total available capacity around 7.5 Gbps. That means that
whenever that home wants to burst to its full 7.5G Tmax, it will need to use every ounce of spectrum
available to it. That would leave practically nothing left for the remaining subs (i.e. Nsub*Tavg
component from the formula).

In reality, the CMTS would not let these other subs starve, but will tend to allocate higher capacity lower
spectrum to the bursting modem and others near the EoL. Meanwhile modems closer to the node and
amplifier outputs will be using their higher capacity profiles in the higher frequency OFDM channels. So,
the bursting modem might only get 5-6 Gbps out of its 7.5G tier during peak busy times.

So, how probable is this worst-case scenario? There are several factors that must all align for this to
happen. These factors and the individual probability of each include:

• Subscriber takes the premium 7.5G DS top billboard tier [~1% to ~5%]
• Subscriber home is on the 5th tap [20%, i.e. 1 out of 5]
• The probability that this link even has 5+ taps [~30%]
o Majority of amplifier outputs have 4 taps or fewer
• The tap is after the 5th or 6th amplifier [~25% of amplifier outputs on N+5/N+6 plant]
• % of MSO plants that are even N+5 or higher [1% to 20%???]
• The amplifier output is being stretched beyond unity gain [~5% to 15%]
• The RG-6 drop cable is >125’ [~25% to 50%]

On N+5/N+6 plants, only 1 out of every ~200 to ~1,000 amplifier outputs might fall into this category.
While it varies greatly by MSO, N+5 and N+6 plants are also becoming a smaller and smaller percentage
of their total HFC plants. Of these potential problem amplifier links, it is only a problem if there is a
subscriber on the 5th tap that takes the 7.5G DS top billboard tier. For all these stars to align, it looks like a
probability on the order of four or five 9’s that this won’t happen.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 31

It may be the end of this decade before the 7.5G DS tier is being offered and this worst-case scenario can
potentially kick in. So, what options are available to the operator over this time window to handle this
potentially rare event? Here are some possibilities:

1. MSO accepts slight QoE degradation (i.e. 7.5G drops to 5-6G during peak busy hour for these
very infrequent customers)
2. MSO only offers 4G-5G DS service tier to this customer
3. Pull fiber to that subscriber’s home and switch them to 10G PON
4. Reduce the N+5/6 cascade depth
5. Add mid-span amplifiers to boost higher frequency signals
6. Replace the RG-6 drop cable with RG-11

Option 1 might be very acceptable, but potentially dependent on your regulatory environment. The
customer perceived QoE between 5 Gbps bursts and 7.5 Gbps burst capacity may not be perceptible. And
this is only happening for rare times when the consumers bursts to the max and for a very small percent of
the customer population. If the operator is still uncomfortable with this, they can choose option 2 and only
provide the consumer with a slightly reduced Tmax (e.g. 5G instead of 7.5G). Note, this is something that
the DSL world has had to always deal with, but this is not nearly as dramatic. In older DSL technology,
customer capacity might have varied from 2 Mbps to 25 Mbps.

If the operator wants to correct the situation and provide additional capacity where needed, then options
3-6 would need to be considered over the next 5-8 years. An economic analysis of some of these choices
is in the next section. Since this scenario does not need to get resolved until the end of the decade, the
operator has plenty of time to correct the issue before it happens.

Everyone agrees that the long-term strategy is to eventually get to FTTH. However, many consider this a
multiple decade transition to get to the point where fiber is pulled down every street. Option 3 jumps
directly to FTTH as the solution. However, since this is a N+5/N+6 plant, chances are that the fiber is not
nearly as close to the home as it might be in N+0/N+1 plant. It might be over a mile away from the
customer. So, option 3 is expected to be the most expensive.

With the long-term strategy to continually push fiber deeper, then option 4 above is making a step in that
direction. If these problem amplifier spans can be identified today (e.g. N+5/N+6 with stretched outputs),
then the operator can spend the next decade gradually attending to these and pushing fiber deeper. Grey
optics aggregator (GOA) / grey optics terminator (GOT) is a cost-effective fiber deep architecture that
could be used to reduce the cascade length, without necessarily going all the way to N+0. The GOA
functionality is placed at the parent node location. The downstream optical signal is split and passed on to
the GOT node using short distance optics. The GOA also aggregates all of the GOT return signals. The
GOT nodes are transparent to the head end and does not require any additional head end optics.

A low cost GOT fiber node can replace one of the multi-bridger amplifiers on the path to these problem
spans to reduce the cascade length. If an operator has a 20 year plan to convert to a fiber deep network
(e.g. 5% of plant per year), they can target these problem spans now to make sure issues with these are
corrected before the end of the decade. Later in time, the GOT node can be upgraded to a full-fledged
fiber node as needed.

Options 5 and 6 both fix the potential capacity issue but are not in alignment with the strategic direction
of pushing fiber deeper towards eventual FTTH. Option 6, pulling a new drop cable, could be aligned
with this strategy if it is a “siamese” cable that contains both fiber and RG-11 drops. Option 5, adding a
mid-span amplifier, might make sense if the logistics do not allow option 4 to pull fiber deeper to reduce
cascade lengths.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 32

7. Cost and Logistic Analysis
Our starting point for economic analysis discussion is a 750 or 860 MHz HFC network case study
depicted in Figure 30. The fiber node area, serving 400 HP, is also a single Integrated Converged Cable
Access Platform (I-CCAP) service group. It is a 42/54 MHz “sub-split” system with a total of 20 RF
amplifiers, 10 each of bridger and line extender type. There is a total of 100 taps, on 25,000 feet of
hardline coax plant, for an average density of 84 HP/mile and ~4.2 amps/mile. The longest RF amp
cascade length is N+4.

Figure 30 – I-CCAP N+4 HFC network with 42/54 MHz sub-split

The cost of upgrading this network to 1218 MHz with a 204/258 MHz high-split is analyzed by (a)
replacing “ePack” node and amps modules and (b) adding appropriate I-CCAP license to augment both
DS and US data capacity. The left-most column of Figure 31 depicts this network upgrade, along with the
percent breakout of various elements. Furthermore, this 1218/204 MHz upgrade is normalized to 100%,
in order to be the baseline to compare to the other cases, namely 1.2 GHz FDX N+0 and 1.8 GHz ESD
N+4. The 2nd column shows the 1218/204 MHz upgrade where the tap faceplates are also upgraded to 1.2
GHz. This adds ~25% premium on top of the base case.

The network shown in Figure 30 is applicable to the 1.2 GHz high-split case. For the other two cases,
however, networks of Figure 32 and Figure 33 are the respective representations. FDX requires a N+0
“fiber-deep” upgrade. The original fiber node plus 20 amplifiers end up being replaced by 4 fiber nodes
with ~50% of the original hardline coax over lashed by new fiber. For the 1.8 GHz ESD N+4 case, whole-
station amplifiers, as well as all 100 tap housings are replaced.

As shown in Figure 31, the FDX N+0 upgrade comes to 454% of the “base case”, that is, the 1.2 GHz
high-split upgrade. The fiber material plus labor costs really dominate for the N+0 upgrade. Our analysis
assumed 80% aerial plant. This component could be much higher if the percentage of underground plant
increases over 20%. The 1.8 GHz ESD upgrade comes to 214% of the “base case”, or a 114% premium.

If only 1 to1.5 Gbps US service tiers are needed, the 1218/204 MHz upgrade is the most cost effective.
However, once multi-Gbps US tiers are needed, then the FDX N+0 or the 1.8 GHz N+4 upgrade is
required. Our study shows that the FDX N+0 upgrade is roughly twice the cost of the 1.8 GHz ESD N+4
upgrade for this N+4 plant case study.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 33

 500%
CAPEX Estimates for Various Network Architecture Upgrades
450%
CMTS
400%
100% = 1.2 GHz High-Split Upgrade

HEO
350%
Node HW
300%
Field HW
250%
Field labor, incl. node
200%
Fiber, mat'l & labor
150%
Taps, mat'l & labor
100%

50%

0%
1218/204 MHz N+4 1218/204 MHz N+4 with 1218 MHz FDX N+0 1794 MHz ESD N+4 Upgrade
without new taps new tap faceplates

Network Architecture

Figure 31 – CAPEX estimates for 1218/204 MHz N+4 (both without and with new tap
faceplates), 1218 MHz FDX N+0, 1794 MHz ESD N+4 upgrades

Figure 32 – DAA N+0 HFC Network Model for 1.2 GHz FDX Upgrade

Figure 33 – DAA N+4 HFC Network Model for 1.8 GHz ESD upgrade

© 2020 SCTE•ISBE and NCTA. All rights reserved. 34

 Figure 34 – Monte-Carlo analysis of “premium” for 1.8 GHz ESD case

There are other benefits to the ESD approach. For one, it is providing dedicated upstream capacity while
FDX is relying on statistical time sharing of US+DS spectrum. The second is that ESD is a much simpler
system than FDX which should result in OPEX savings through reduced maintenance and easier to
diagnose plant issues.

Because many elements go into an evaluation of this type, a Monte-Carlo analysis was run with various
assumptions varying over a reasonable range for the ESD upgrade case. A total of 100,000 Monte-Carlo
samples were run to see how much the ESD “premium” might vary. Figure 34 shows that the ESD
“expected” premium of 114% expands into a 94% to 134% “confidence interval”.

What is driving the 114% ESD premium, while the FDX N+0 premium is 354%? Figure 31 offers some
answers: fiber over lash required for N+0 accounts for 264% of the 354% premium for FDX! The 2nd
most dominant FDX component is node hardware at ~90%. For ESD, RF amplifiers, labor to replace
those, and a material and labor to upgrade tap housings and faceplates to 1.8 GHz account for most of the
cost.

For the ESD case, a reasonable “workload” for the whole amplifier and taps upgrade were considered,
especially allowing for the whole housing to get replaced. This assumed there is 1 tap per 4 HP; the RF
amplifiers are closer to 1 to 20 HP ratio, and nodes are even further – in the assumptions made above it’s
1 to 400 HP. If an amplifier module takes approximately 1 hour to replace and a tap housing takes 45
minutes to replace, how much of an additional labor force would an operator need in order to do these
upgrades?

For an MSO serving a region with population of one million people (e.g. ~400,000 HP), Table 4 holds the
answer. For the taps portion, it would require an additional full-time crew of ~10, working full-time, over
a 5-year period; and doing nothing else. One may describe it as a large task, but 10 additional employees
to do this task seems reasonable.

The previous section discussed a corner case where a subscriber with a long RG-6 drop cable at the EoL
on a N+5/N+6 stretch plant does not have sufficient network capacity to obtain the highest service tier. Of
the half dozen potential options mentioned, the most cost effective would be to replace the drop cable
with RG-11. The operator might also consider putting it in a conduit for reliability and ease of later
upgrading to a fiber drop as well as consider siamese cable/fiber pair. Adding a mid-span boost amplifier
for high frequencies on the last link might triple or quadruple the cost of the drop cable option. And, more

© 2020 SCTE•ISBE and NCTA. All rights reserved. 35

than one mid-span boost amplifiers might be needed on some of the other links in the cascade. Neither of
these options pushes fiber deeper so they may not be aligned with the longer strategic direction.

Table 4 – Estimate of Labor force required to replace amps and taps in 5 years
Nodes RF Amps Taps Homes-Passed
Quantity 1,000 20,000 100,000 400,000
Task duration 1 Hour 1 Hour 45 minutes
24/7 Person-Years 0.1 2.3 8.6
40hr/week Person-Years 0.5 9.6 36.1
30hr/week Person-Years 0.6 12.8 48.1
Required size crew to
complete the task in 5 years
<<1 <3 <10

Trying to jump directly to FTTH can be extremely costly. Since this home is EoL on a N+5/N+6 plant, it
might be up to a mile and a half from the fiber node. The costs of pulling fiber can skyrocket, especially if
the plant is mostly underground. Some estimates show this to be 100 times the cost of just replacing the
drop cable. This is hard to justify for a single customer.

One strategy is to push fiber deeper over a 20-year window to get most customers within 1000’ or 1500’
of fiber access. At this point, the operator can offer FTTH on demand when and even if needed (e.g.
90%+ of subs may stay on cable ‘forever’). To achieve this, the operator can focus on the N+5/N+6
stretch plants in the near-term, with the goal of getting all of their HFC down to N+2/N+3 by the end of
the 2020 decade. Then in the decade that follows, the operator can push fiber deeper to achieve N+0/N+1
that can enable FTTH on demand by the end of the 2030 decade. Analysis such as [ULM_2016] have
shown that fiber deep N+0/N+small upgrades are more cost effective than FTTH for HFC brownfield
upgrades.

8. Conclusion
The title of the paper posed the question “Is “Unity Gain” Still the #1 Objective?” as the cable world
migrates to 1.8 GHz DOCSIS 4.0 plant. The paper has shown how much more difficult the 1.8 GHz unity
gain task will be compared to 1218 MHz. Looking at stretch plant, the unity gain starts to come up short
and the cracks become obvious at the longer cascades like N+5 and N+6.

Previously, conventional wisdom would have said that the amplifiers would need to be re-spaced; a
potentially expensive proposition. Recently, there has been discussions of adding a mid-span amplifier to
boost the gain for these links, which should not be taken lightly, as this adds more active components into
the system.

Instead, inherent DOCSIS capabilities can leveraged to compliment unity gain when it comes up short.
Previous thoughts of using EoL throughput as an indicator of plant capacity are no longer accurate at 1.8
GHz. DOCSIS 3.1/4.0 OFDM channels with variable bit-loading and multiple profiles enable the CMTS
scheduler to maximize the system capacity. This was shown nicely in Figures 18 & 20, which are
repeated in Figure 35 below. The N+6 stretch plant saw DOCSIS weighted average capacity gains that
were 35% higher than EoL throughput.

The paper explored some of the variables that impact system capacity. In addition to amplifier spacings
(e.g. typical vs. stretch plants), it looked at various cable drop lengths and the addition of super-stretched

© 2020 SCTE•ISBE and NCTA. All rights reserved. 36

links into a typical plant. These are all scenarios that allow DOCSIS to optimize the system capacity some
more.

The 1.8 GHz ESD plant offers several options for different upstream splits, varying all the way up to
684/834 MHz. Choosing different splits lets the operator balance between upstream and downstream
bandwidth. The paper looked at the DS capacities associated with the different splits and then showed the
range of DS + US service tier SLA combinations that might be supported near the end of this decade.

492-1794 MHz Throughput vs Tap Position in N+6 Stretch Plant, 150' Drop cables End-of-Line vs. Weighted Avg Throughput - 492-1794 MHz, 150' Drop cables
12
12

60%
10 11

EoL EoL EoL

Net Throughput [Gbps]

8
N+1 EoL
Throughput [Gbps]

N+0 N+2 EoL 10

N+3 EoL EoL
N+4 N+5
6 N+6
9

8
2

0 7
0_1 0_2 0_3 0_4 0_5 1_1 1_2 1_3 1_4 1_5 2_1 2_2 2_3 2_4 2_5 3_1 3_2 3_3 3_4 3_5 4_1 4_2 4_3 4_4 4_5 5_1 5_2 5_3 5_4 5_5 6_1 6_2 6_3 6_4 6_5 6_6 Node+0 Node+1 Node+2 Node+3 Node+4 Node+5 Node+6
Tap Number Cascade

Stretch Plant 492-1794 Typical EoL 492-1794 Typical Avg 492-1794 Stretch EoL 492-1794 Stretch Avg

Figure 35 – EoL vs. Weighted Avg Throughput – Typical and Stretch Plants

The cost and logistic analysis section hopefully gave the reader some sense of the economic tradeoffs that
will be encountered when looking at these various options. When the operator finally needs multi-Gbps
US tiers, the 1.8 GHz ESD is more cost effective, potentially half of the cost of a FDX N+0 upgrade.

So, to answer the question from our title, the answer is: Maybe Yes! The authors still believe that trying
to maintain unity gain is a key objective in any HFC design. It has shown its worth time and time again.
However, as it approaches the breaking point, DOCSIS scheduling can fill the gaps on many of these
plants on the bubble, reducing the number of HFC links that need drastic action to a significantly smaller
amount. The two work well together.

Acknowledgements
The authors would like to thank Cox Communications and in particular David Job, Principal Eng of OSP
Engineering at Cox, for providing the data and insights into the HFC amplifier spacing distribution.

The authors are also extremely grateful to everyone that gave us insights and advice on HFC design and
doing a sanity check on what “reasonable” might be. In particular, we want to thank Stuart Eastman of
CommScope for sharing some of network design wisdom; and Jay Lazorcik of CommScope for HFC
network and unity gain diagrams.

© 2020 SCTE•ISBE and NCTA. All rights reserved. 37

 Bibliography & References
[ALB_2019] A. Al-Banna et. al., “Operational Considerations and Configurations for FDX & Soft-FDD,”
SCTE Cable-Tec 2019, SCTE

[CLO_2019] T. J. Cloonan et. al., “Capacity Planning, Traffic Engineering, And HFC Plant Evolution For
The Next 25 Years,” SCTE Cable-Tec 2019, SCTE

[CLO_2017] T. J. Cloonan et. al., “The Big Network Changes Coming with 1+ Gbps Service
Environments of the Future,” SCTE Cable-Tec 2017, SCTE

[CLO_2016] T. J. Cloonan et. al., “Using DOCSIS to Meet the Larger BW Demand of the 2020 Decade
and Beyond,” NCTA Spring Technical Forum 2016, NCTA

[CLO_2014] T. J. Cloonan et. al., “Simulating the Impact of QoE on Per-Service Group HSD Bandwidth
Capacity Requirements,” SCTE Cable-Tec 2014, SCTE

[CLO_2013] T. J. Cloonan et. al., “Advanced Quality of Experience Monitoring Techniques for a New
Generation of Traffic Types Carried by DOCSIS,” NCTA Spring Technical Forum 2013, NCTA

[EMM_2014] “Nielson’s Law vs. Nielson TV Viewership for Network Capacity Planning,” Mike
Emmendorfer, Tom Cloonan; The NCTA Cable Show Spring Technical Forum, April, 2014

[FDX_PHY] “DOCSIS 4.0 Physical Layer Specification”, CM-SP-PHYv4.0-D01-190628, CableLabs

2019

[FDX_XSD_IBC] “Full duplex DOCSIS & Extended Spectrum DOCSIS Hold Hands to Form the 10G
Cable Network of the Future”, by F. O’Keeffe et. al., IBC 2019

[ULM_2019] J. Ulm, T. J. Cloonan, “The Broadband Network Evolution continues – How do we get to
Cable 10G?”, SCTE Cable-Tec 2019, SCTE

[ULM_2018] J. Ulm, “Making room for D3.1 & FDX – Leveraging Something Old that is New Again!”,
SCTE Journal of Network Operations : Find Fresh Approaches to Plant-Related Topics, Vol 4. No. 1. Dec
2018, SCTE

[ULM_2017] J. Ulm, T. J. Cloonan, “Traffic Engineering in a Fiber Deep Gigabit World”, SCTE Cable-
Tec 2019, SCTE

[ULM_2016] “Adding the Right Amount of Fiber to Your HFC Diet: A Case Study on HFC to FTTx
Migration Strategies”, John Ulm, Zoran Maricevic; 2016 SCTE Cable-Tec Expo

[ULM_2014] “Is Nielsen Ready to Retire? Latest Developments in Bandwidth Capacity Planning”, John
Ulm, T. Cloonan, M. Emmendorfer, J. Finkelstein, JP Fioroni; 2014 SCTE Cable-Tec Expo

© 2020 SCTE•ISBE and NCTA. All rights reserved. 38

 Abbreviations
BW bandwidth
CAGR compounded annual growth rate
CAPEX capital expense
CCAP Converged Cable Access Platform
CDF cumulative distribution function
CM cable modem
CMTS Cable Modem Termination System
CPE consumer premises equipment
D3.1 Data Over Cable Service Interface Specification 3.1
D4.0 Data Over Cable Service Interface Specification 4.0
DAA distributed access architecture
DOCSIS Data Over Cable Service Interface Specification
DS downstream
EOL end of line
EPON Ethernet Passive Optical Network (aka GE-PON)
ESD extended spectrum DOCSIS
FDX full duplex (i.e. DOCSIS)
FTTH fiber to the home
FTTx fiber to the ‘x’ where ‘x’ can be any of the above
Gbps gigabit per second
GHz gigahertz
GOA grey optics aggregator
GOT grey optics terminator
HEO head end optics
HFC hybrid fiber-coax
HP homes passed
HW hardware
I-CCAP Integrated Converged Cable Access Platform
LDPC low density parity check (FEC code)
LE line extender
MAC media access control
MB multi-port bridger
Mbps megabit per second
MHz megahertz
MSO multiple system operator
N+0 node+0 actives
NCTA The Internet & Television Association
OFDM orthogonal frequency-division multiplexing
OPEX operating expense
PHY physical interface
PON passive optical network
PSD power spectral density
QAM quadrature amplitude modulation
QoE quality of experience

© 2020 SCTE•ISBE and NCTA. All rights reserved. 39

RF radio frequency
SG service group
SCTE Society of Cable Telecommunications Engineers
SNR signal to noise ratio
STB set top box
Tavg average bandwidth per subscriber
TCP total composite power
Tmax maximum sustained traffic rate – DOCSIS Service Flow parameter
TX transmit
US upstream

© 2020 SCTE•ISBE and NCTA. All rights reserved. 40