See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251953241

A versatile multichannel filter bank with multiple channel bandwidths

Article · January 2010

DOI: 10.4108/ICST.CROWNCOM2010.9238

CITATIONS READS
5 938

3 authors:

fred joel harris Robert Mcgwier

San Diego State University Virginia Polytechnic Institute and State University
189 PUBLICATIONS 7,845 CITATIONS 37 PUBLICATIONS 506 CITATIONS

SEE PROFILE SEE PROFILE

B. Egg

10 PUBLICATIONS 29 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cybersecurity Work View project

Variable Star Astronomy View project

All content following this page was uploaded by B. Egg on 04 August 2015.

The user has requested enhancement of the downloaded file.

 A Versatile Multichannel Filter Bank with Multiple Channel Bandwidths
fred harris Robert McGwier Benjamin Egg
San Diego State University & CCR-Princeton CCR-Princeton San Diego State University

ABSTRACT: lection of the number of Nyquist zones to be separated by the

The FFT is an important core to which we couple an channelizer. Here the number of zones or channels is defined
M-path partitioned filter and commutator to build high per- by M, the size of the IFFT. The M in the M-point transform
formance spectrum analyzers and channelized transmitters defines
and receivers. With slight modifications of the system the
channelizer and spectrum analyzer can perform secondary Polyphase
Partition
processing tasks such as arbitrary sample rate changes. Out- fs
H0(Z)
put sample rates can be adjusted to be a multiple of channel
spacing or a multiple of the channel symbol rate. With minor H1(Z)
additional processing the channelizer can also coherently
combine signals from adjacent frequency bands to synthesize FDM H2(Z)

arbitrary bandwidth super channels from standard prototype H3(Z) M-PNT

IFFT
TDM

.....
. . .. .
.....
channels. M fs

I. INTRODUCTION: HM-2(Z)
The most common form of an M-path channelizer per- HM-1(Z)
forms M-to-1 down-sampling of an input series by delivering
M-successive input samples to the input ports of an M-path Hr (Z)=Z{h(r+nM)}
partitioned low-pass filter. Each output port of the M-path
filter contains an aliased time signal caused by the input Figure 1. M-to-1 Down-Sample, M-Path Polyphase Channelizer.
down-sampling. In the standard channelizer the aliases are
the M-to-1 spectral folds of the Nyquist zones formerly cen- both the width of the successive Nyquist zones and the spac-
tered on the M-multiples of the output sample rate. ing between channel spectral centers as fS/M, where fS is the
The alias components in each path have different and input sample rate. The second task performed by the channe-
unique phase profiles partly due to the time delays induced by lizer is the definition of the channelizer spectral response
the commutator process and partly due to the phase shift which includes pass-band and stop-band boundaries as well as
response of each path in the M-path filter. When phase rota- pass-band and stop-band ripple. The channel spectra match
tors matching the k-th multiple of the M-roots of unity are the aliased spectrum of the prototype low-pass from which
applied to the output time series from each path, the phases the M-path filter was partitioned. The third task is that of re-
of the aliased k-th Nyquist zone in each path filter are aligned. sampling from the input sample rate of fS to the output chan-
When the phase aligned alias terms are added they form a nel sample rate of fS/M. It is obvious this process occurs in the
coherent sum of that spectral component. On the other hand, commutator which delivers M input sample to the polyphase
the phase profiles of the remaining Nyquist zones are aligned system for it to compute 1 output sample from each channel.
with the M-roots of unity and are destructively canceled when Since three different processes are responsible for the chan-
summed. Intuitively, the extraction of the time series from nelizer parameters of channel spacing, channel bandwidth,
any selected aliased Nyquist zone is possible because the time and channel sample rate we can independently select and
series from each path supplies one of the M-equations re- adjust them to obtain useful variations of the basic channeliz-
quired to solve for the M-unknowns, the aliased components er.
from the M Nyquist zones.
When the time signals residing in multiple aliased Nyquist II. CHANNELIZER PARAMETERS.
zones are to be separated from the M-fold aliased baseband In the previous section we commented that the three pa-
signal, the collection of phase rotators and coherent sums is rameters of channel spacing, channel bandwidth, and channel
most efficiently applied to the output of the M-path filter by sample rate could be independently selected and controlled.
the inverse fast Fourier transform (IFFT). This common struc- Figure 2 shows a set of options illustration possible relation-
ture of the M-path down-sampler and channelizer is shown in ships between these parameters. This set does not exhaust
figure 1. the list of possible options. Note that in the 4 cases shown
We note that the M-path polyphase channelizer performs here, the channel spacing is the same and equal to fS/M. This
three distinct operations and that these operations occur in tells us that the IFFT is of length M and that the filter is likely
different segment of the channelizer. The first task is the se- an M-path filter.
 The channel bandwidth of the first option is less than the
channel spacing while the channel bandwidth of the second In the fourth option, the output sample rate is twice the
option is equal to the channel spacing. The filter bandwidth is channel spacing. The advantage of this option is two-fold. First
controlled and defined in the design of the low pass prototype it satisfies the Nyquist sampling criteria of the channelized
filter that is partitioned to become the M-path structure. The signals for signal bandwidths less than or equal to the channel
first filter option would likely be selected for a communication spacing and second it avoids the spectral folding at the chan-
receiver that needs to separate adjacent spectral bands while nel band edge for channel widths equal to the channel spac-
the second filter ing. This desired property is illustrated in figure 3.

f BW = 0.8 fs/M f =fs/M III. M/2 Down-Sample in Polyphase Filter Bank.

We now modify the M-path polyphase filter to perform
the sample rate change from the input rate fS to the output
f
rate 2fS/M. We have presented one form of this modification
f SMPL =fs/M
in an earlier paper but we choose to present the derivation
f BW=fs/M f =fs/M
here in a slightly different form to be consistent with the dual
form we present in a later section of this paper. We develop
f and illustrate the modification with the aid of Figures 4a
f SMPL =fs/M through 4d. Figure 4a presents the structure of the M-path
f BW = 0.8 fs/M f =fs/M filter implementation of the polyphase partition shown in
eqn. 1 for the specific M-path filter partition.

f N-1 M-1 (N/M)-1

f SMPL =2(0.8fs/M) H(z) = h(n)z -n = h(r + nM)z -(r+nM)
f BW=fs/M f =fs/M n=0 r=0 n=0
M-1 (N/M)-1 M-1
= z -r h(r + nM)z -nM = z -rHr (zM ) (1)
f r=0 n=0 r=0
f SMPL =2(fs/M) (N/M)-1
where Hr (zM ) = h(r + nM)z -nM
Figure 2. Some Channelizer Parameters Relating Channel Spacing, n=0
Channel Bandwidth, and Channel Sample Rate.
Note the M/2-to-1 rather than the conventional M-to-1 down-
option would likely be selected for a spectrum analyzer which sample operation after the output summing junction. In figure
must monitor and report the energy content in contiguous 4b we apply the noble identity [1] to the polyphase paths and
gap free spectral intervals. In these first two options the sam- pull the M/2-to-1 down-sampler through the path filters
ple is shown to be fS/M telling us that this is a maximally de- which convert the polynomials in ZM operating at the high
cimated filter bank with M-input samples for 1-outpt sample. input rate to polynomials in Z2 operating at the lower output
The third and fourth options shown in figure 2 match the rate. Note the paths are now polynomials in Z2 rather than
channel spacing and channel bandwidths of the first two op- polynomials in Z as is the normal mode that we identify in the
tions but differ in output sample rates. In the third option, the maximally decimated filter bank. Figure 4c shows the second
output sample rate is equal to twice the channel symbol rate, application of the noble identity in which we again take the
a rate greater than the channel spacing. This option is desired M/2-to-1 down-sampler through the Z-M/2 parts of the input
for communication receivers that like to perform the syn- path delays for the paths in the second or bottom half of the
chronization and equalizations tasks required for signal de- path set. In figure 4d the M/2-to-1 down-sampling switches
modulation at 2-samples per symbol. and their delays are replaced with a two pronged commutator
BW BW
that deliver the same sample values to path inputs with the
same path delay. Here we also merged the Z-1 delays in the
Aliased
Transition lower half of filter bank with their path filters. Figure 5 shows
Bandwidth and compares the block diagrams of the path filters in the
f f upper and lower half of this modified polyphase partition.

Channel Spacing
The final modification to the polyphase channelizer is the
Channel Spacing
Sample Rate =Channel Spacing Sample Rate=2 Channel Spacing time alignment of the shifting time origin of the input samples
in the M-path filter with the stationary time origin of the
Figure 3. Channel Spectral Folding when Sample Rate Equal to phase rotator outputs of the IFFT. We can understand the
Channel Spacing and No Spectral Folding when Sample Rate Equal problem by visualizing a single cycle of a sine wave extending
to Twice Channel Spacing. over M samples being inserted in the input data register, the
first column of the polyphase filter in segments of length M/2. H0( Z )
2

We can assume that the data in the first M/2 addresses is

2
phase aligned with the first M/2 samples of a single cycle of H1( Z )
the sine wave offered by the IFFT. This is shown in figure 6.

....
2
H M -1( Z )
M 2
H0( Z )
-1 2
M Z HM ( Z )
-1 H1( Z ) 2
Z
-1 2
....

....
Z H M +1( Z )
2

-( M -1) M

....
Z 2 H M -1( Z )
2

-1 2
-( M ) HM( Z
M
) M Z HM-1 ( Z )
Z 2
2 :1
2

M
-( M +1) H M +1( Z )
Z 2
2
Figure 4d. Commutator Replaces Path Delays
....
....

2
Hk (Z )
M
-(M-1) HM-1 ( Z )
Z

Figure 4a. M-Path Filter and M/2 Down-Sample h(k) h(k+M) h(k+2M) h(k+3M) h(k+4M)

2 -1 2
H0( Z ) Z H(k+M/2)(Z )
M :1
2
2
-1 H1( Z )
Z
M :1
2
....

....

h(k+M/2) h(k+3M/2) h(k+5M/2) h(k+7M/2) h(k+9M/2)

2
-( M -1) H M -1( Z )
Z 2
2
M :1
2
-( M ) HM ( Z )
2 Figure 5. Path Filters with and without Extra Delay
Z 2
2
M :1
2 2 When the second M/2 input samples are delivered to the
-( M +1) H M +1( Z )
Z 2
2
M :1 input data register the first M/2 input samples shift to the
2
....
....

second half of the M-length array. Its original origin is now at

-(M-1) HM-1 ( Z )
2 address M/2 but the IFFT’s origin still resides at address 0. The
Z
M :1 origin shift between the two sine waves causes the input sine
2
wave in the register to have the opposing phase of the sine
Figure 4b. Apply Noble Identity to Path Filters wave formed by the IFFT. What we are observing is the sinu-
soids with an odd number of cycles in the length M array alias
H0( Z )
2 to the half sample rate when down sampled M/2-to-1. We
M :1
2 can respond to this by phase reversing the odd indexed IFFT
2
Z
-1 H1( Z ) sinusoids on alternate outputs. Alternatively, knowing that
M :1
phase shift and time delay are equivalent for a sine wave we
....

2
....

....

-( M -1) H M -1( Z )
2 can perform M/2 point circular shifts of alternate M-length
Z 2
2
M :1
2
vectors formed by the polyphase filter before presenting the
2
HM( Z )
2
vector to the IFFT. The circular shifts that perform the re-
M :1
2 2 quired phase reversals of successive input vectors presented
-1 -1 HM (Z )
Z Z +1
M :1 2 to the IFFT are illustrated in figure 7.
2
....

....

....

Z
-(M -1)
2 Z
-1 HM-1 ( Z )
2
IV. CHANNELIZER PERFORMANCE.
M :1
2 We designed and simulated a 128 path polyphase chan-
nelizer with a prototype 1536 tap low pass filter. The parti-
Figure 4c. Apply Noble Identity to Delays in Path Delays tioned 128 path filter contains 12 taps per path. The 128-path
filter operates as a 64-to-1 down sampling channelizer. If we
hypothesize a 128 MHz input sample rate, then the channel
 Length M Co lumn of Polyphase Partition

First M/2 Input Samples power spectrum. Figure 10 is a log magnitude plot of signal
1
variance of each channel versus channel index. Note from the
data origin markers on this plot that the signal bandwidth of the 2-MHz
bands are spanned by 3-channel bands while the 1-Mhz signal
FFT origin FFT origin
Length M Co lumn of Polyphase Partition band is spanned by a single channel filter and the 12 MHz
Second M/2 Input Samples
signal band is spanned by 11 channel bands.
2 1

data origin
V. PARTITIONING AND COMBINING CHANNEL BANDWIDTH.
FFT origin FFT origin The channelizer we presented in the previous section
Length M Co lumn of Polyphase Partition

Circular Shift
outputs complex time series from 1-MHz wide channels at 2-
1 2 MHz sample rate. If we have need for finer resolution channel
data origin portioning we simply deliver the complex time series to a
second tier channelizer which can further partition the se-
FFT origin FFT origin
lected band. The second tier channelizer is a replica of the
Figure 6. Illustrating Phase Reversal of M-Point Sinusoid Input to first tier channelizer. It is implemented as an M 2-path poly-
M/2 Path-Polyphase Filter. phase filter with an IFFT of length M2 which interact to chan-
nelize and down sample M2-to-1. As an example, selecting M2
0 0 to be 16, will process the 1-MHz bandwidth channel signal
1 1
sampled at 2-MHz to form 62.5 kHz sub channels at 125 kHz
M-Path Input Data Buffer

Circular Output Buffer

M-Path Polyphase Filter

2 2
FDM
3 sample rate.
....

M-Point IFFT

M/2-1 M M M
........

M/2 128 Channel Channel Filter Frequency Response: 6 dB Crossover

M/2+1
TDM 0
Log Magnitude (dB)
….

M-2
M-1 M-1 -50

State Engine -100

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Normalized Frequency, (f/f CH)
Figure 7. M/2 Down-Sampled M-path Polyphase Channelizer
Figure 8. Frequency Response of Three Adjacent Channel Bands
Spectra at Output of Modulator
spacing of the channelizer is 1 MHz and the sample rate per 0
channel is 2 MHz. The prototype filter is designed so that
Log Magnitude

-20

adjacent channel filters cross at their -6 db level. This means

(dB)

-40

the filter is a Nyquist filter and broader bandwidth channels -60

-80
can be synthesized by simple summations of the adjacent
-100
channel signal components. We describe this process in the -120
-40 -30 -20 -10 0 10 20 30 40
next section. The frequency response of the channels in the Channelizer Channel Index
0 1 2 5 6 1
7
channelized receiver is shown in figure 8. Note the 100 dB 1
0
1
0
1
0
1
0
3 1
0
4 1
0
1
0 0
-1 -1 -1 -1
-1
dynamic range of the filter design. -1 0 1 -1 0 1
-1
-1 0 1
-1
-1 0 1
-1
-1 0 1 -1 0 1 -1 0 1 -1 0 1

1
8 1 1 1 1 1 1
15
To demonstrate the channelizer performance we de- 0
-1
0
-1
9 0
10 0
11 1
0 12 0
-1
13 0 14 0
-1
-1 -1 -1
-1
signed and simulated a modulator that forms multiple signal -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1
-1 0 1 -1 0 1
1
-1 0 1
1
16 1
17 1
18
1
19
1
20
1
21
1
22 23
channels with 4-MHz channel spacing. Twelve of its sixteen 0
-1
0
-1
-1
0 0
-1
0
-1
0
-1
0
-1
0
-1

channels contain QPSK signals with 2 MHz symbol rates and 3- -1 0 1 -1 0 1

25
-1 0 1
26 1
-1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1
1
24 1 1
27 1
28 1
29 1
30 1
31
MHz bandwidth. One channel contains a narrow band QPSK 0
-1
0
-1 -1
0 0
-1
0
-1
0
-1
0
-1
0
-1

signal with 0.5 MHz symbol rate and 1-MHz bandwidth. One -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1

set of three channel bands share a wideband QPSK signal with Figure 9. Input Spectrum and Channel Complex Output Signal
8-MHz symbol rate and 12 MHz bandwidth. The spectrum of from 32 Positive Index Channels of 128 Channel Channelizer
the modulation test signal is shown in the upper half of figure
9. The lower half of figure 9 shows the complex signal trajec- We quickly see that we have easy access to enhanced
tories from the positive frequency indices 0-to-31 of the 128 resolution channelization of any of the time series formed by
channel channelizer. The variance or spread of these trajecto- the base-banded, filtered and down-sampled time series ob-
ries is proportional to the energy content in that channelized tained from the first tier channelizer. In one option, we can
channel. Note that channels 1, 2, and 3 cover the 2-MHz band cover each output channel of the first spectral partition with a
centered at 1 MHz and channels 9 through 19 cover the 12- second tier channelizer to partition the entire spectrum into
MHz band centered at 14 MHz. A plot of signal variance ver- higher resolution channels. In another option, we can have a
sus channel index is a valid representation of the input signal background spectral sniffer identify channels containing sig-
 nals of interest and direct a second tier channelizer to parti-
tion selected channels.

Short Term Power Spectra from 128 Channel Channelizer

0
Log Magnitude
(dB)

-50

-100

-60 -40 -20 0 20 40 60

Channel Index
Figure 10. Channel Time Series Variance versus Channel Index. Figure 12. Spectrum Formed from Channelizer and Synthesized
Super Channel
The signals we presented to the channelizer in the pre-
vious section had bandwidths which were wider than the The second tier offers the option of further spectral partition
channelizer filter bandwidth. We have already addressed the or of spectral merging with 16-path polyphase channelizers.
task of partitioning a selected band into narrower bands. We As shown the system is capable of offering three different
now address the dual task of combining the outputs of conti- levels of spectral partition. These second tier processing
guous narrowband channels to synthesize wider bandwidths. blocks can be assigned dynamically to spectral regions that
Remarkably this is a trivial task. The background sniffer or the require the additional partitioning or merging. A number of
spectral estimate obtained from the first tier channelizer can different length second tier channelizers can be used to widen
easily identify wide bandwidth input signals which are the range of available bandwidths.
spanned by a number of the narrow bandwidth channelizer 1-to-8 Up-Sampler
bands. All the time series from the identified bands have been 2 16
in 16-Path Channelizer
MHz MHz
down-sampled and translated to baseband by the channeliz- 64-to-1 Down-Sampler

Cir cula r Buffer

2 16

Post Processing Channel Selector

16

..
ing process. To reassemble the original wider bandwidth sig- in 128-Path Channelizer MHz
Path

..
..
Pnt

..
nal we have to up-sample each time series and translate each 128
IFFT Filter
MHz

..
spectral region to its proper offset frequency and simple add
Cir cula r Buffer
..

128 128
their signal components. This we note is the dual operation of
....
....

....
Path Pnt 2
Filter MHz
the analysis receiver channelizer and this dual task is per- IFFT 125
.. ..

kHz
formed by the dual synthesis channelizer. The dual channeliz- 2

Cir cula r Buffer

MHz
er, formed by an M3 point IFFT and an M3 path polyphase 16 16

..
Path

..
Pnt

..
..
filter performs an M3/2-to-1 up-sampling function as it com-
Filter IFFT
bines the time series from selected adjacent bands output
..

from the first tier channelizer. The form of the dual channeliz- 8-to-1 Down-Sampler
er or combiner is shown in figure 11. in 16-Path Channelizer

0 0 Figure 13. Block Diagram of Two Tier Channelizer with Second

1
1 FDM Tier Channel Combiners and Channel Partitioners
M-Path Input Data Buffer
Circular Output Buffer

M-Path Polyphase Filter

2 2
3
....

REFERENCES:
M-Point IFFT

M/2-1
........

1. fred harris, “Multirate Signal Processing for Communication

M M M
M/2
TDM M/2+1 Systems”, Prentice-Hall, 2004.
….

2. fred harris, “Dual Mode Receiver for Composite UTRA and

M-2
M-1 M-1 CDMA-2000 3G Wireless Signals”, Wireless Personal Mobile
Communications (WPMC-2009), 7-10 September 2009, Sen-
State Engine
dai, Japan
3. fred harris and Chris Dick, “Performing Simultaneous Arbitrary
Spectral Translation and Sample Rate Change in Polyphase In-
Figure 11. Channel Combiner Synthesizes Wide BW Channels terpolating or Decimating Filters in Transmitters and Receiv-
ers”, Software Defined Radio Technical Conference, SDR’02,
Figure 12 shows the spectrum obtained from the time series 11-12 November 2002, San Diego, CA.
output by the super channel formed from the first tier filter 4. fred harris, Chris Dick, Xiaofei Chan, and Elletra Venosa, “M-
bins 9-to-19. Figure 13 shows the block diagram of the pro- path Channelizer with Arbitrary Center Frequency Assign-
posed two tier channelizer. The first tier is the 128 path initial ments”, Wireless Personal Mobile Communication (WPMC-
channelizer that forms multiple 1-MHz wide channels sampled 2010), Recife, Brazil, 11-14 October 2010.
at 2.0 MHz.

View publication stats