ECE 265A – Winter 2019

Lecture 15
Receiver Architecture (2)
Direct-conversion Rx (a)
Vincent Leung

© Leung, ECE265A Winter 2019 1

 Outline

 A high-level discussion of mixers

o Trivia: freq-translation and “image”
o Conversion gain
o SSB and DSB NF
o (noise folding and mixing spurs) – future topics
 Direct-conversion receivers
o LO leakage
o DC offsets
o (IIP3, flicker noise, IQ mismatches) – to be discussed

Ref: Razavi text: page 179-187

© Leung, ECE265A Winter 2019 2

 Mixer Trivia

 Mixer performs frequency translation (by multiplying two signals)

 On Rx (Tx), mixer performs frequency down (up)-conversion
 Mixers have 3 ports: RF, IF* and LO
o IF for heterodyne Rx/Tx; or baseband for homodyne Rx/Tx

LO2 LO1

 LTI circuits cannot produce

freq-conversion. Mixers are either*: Rx
o Time-varying circuits LO1 IF/ baseband*
RF
o Non-linear circuits
Tx
* See 2 examples later on page 11-12
© Leung, ECE265A Winter 2019 LO2 3
 Mixer revisit (1): Freq Translation

 Ideal mixer translates modulated signal from one carrier to another.

Freq
fIF f LO f RF

o Suppose that the input of the mixer are the RF and the LO signals
•
•
o The output is*:
•
o 𝑹𝑭 𝑳𝑶

o The modulated signal is translated to 2 new freqs:

• , and 𝑹𝑭 𝑳𝑶 .
o One of the outputs will be filtered (and discarded).
*Recall: cos𝛼 cos𝛽 = cos 𝛼 + 𝛽 + cos 𝛼 − 𝛽
© Leung, ECE265A Winter 2019 4
Note a heterodyne receiver is assumed in this discussion
 Mixer revisit (2): Image problem

 Note that the LO can be below the RF or above the RF (heterodyne

receiver):
Low-side injection: 𝜔 <𝜔 High-side injection: 𝜔 >𝜔

𝜔 𝜔 𝜔 𝑓𝑟𝑒𝑞 𝜔 𝜔 𝜔 𝑓𝑟𝑒𝑞

 That also means, for any given , two frequency components at

“ are down converted to the same frequency:
o Image problem*
o it is bad because the interferences (and noise) at can
potentially overwhelm the receiver

* It can be dealt with by image-reject

filter, or by “image-reject architectures” (Hartley,
or Weaver Rx) – to be discussed next week
𝜔 𝜔 𝜔 𝜔 𝑓𝑟𝑒𝑞
© Leung, ECE265A Winter 2019 5
 Mixer Conversion Gain

 Defining Mixer “Conversion” Gain:

o voltage conversion gain
 Note that input/ output are at different frequencies (conversion gain)
o In modern RF design, we prefer to employ “voltage (or current)
quantities” (in stead of power, as in “power conversion gain”). It is
because the impedances are often not matched (and mostly
imaginary), making the use of power quantities difficult and
unnecessary.
 Take the ideal multiplier as an example, the mixer gain would be:
o
o 𝑹𝑭 𝑳𝑶

o voltage conversion gain

 On real mixers, the gain is actually more dependent on topology
(passive or active), but less dependent on LO signal amplitudes.
o Mixers operate as switches (or switching amplifier*)
*Like what we discussed in Lecture 12, page 12-13. Passive and
© Leung, ECE265A Winter 2019 active mixers will be discussed in ECE265B, so stay tuned! 6
 Ideal Mixer NF (SSB)

 A “noiseless” mixer should have NF = 0dB, right? No!

 Heterodyne receiver
o Noise in the signal band and the image band are both translated to
by mixer gain ( ). Therefore, for a noiseless mixer:
o
 NF of a noiseless mixer = 3dB
 This is called Single-Sideband (SSB) NF
o This means the signal only exists on “one side” of the LO
(while the noise exists equally on “both sides” of the LO)

© Leung, ECE265A Winter 2019 See Razavi text, page 343-346 7

 Ideal Mixer NF (DSB)

 Homodyne receiver
o Only the noise in the signal band is translated to baseband
 If mixer is noiseless, the mixer input noise = mixer output noise.
Therefore,
o
 NF of a noiseless mixer = 0dB
 This is called Double-Sideband (DSB) NF
o This means the signal is on “both sides” of the LO
(And so is the noise)

© Leung, ECE265A Winter 2019 See Razavi text, page 343-346 8

 General SSB Mixer NF

 Now, assume the mixer is noisy and has non-unity gain.

 Define:
o = mixer conversion gain (assumed equal* on & )
o , = source noise on the RF band
o , = source noise on the image band
o = mixer input-referred noise (assumed equal* on & )

 SSB NF is given by:

, , 𝑵𝒎𝒊𝒙
o
,

, ,
o
, ,
𝑵𝑺,𝑹𝑭 𝑵𝑺,𝑰𝑴

* These are good and practical assumptions

© Leung, ECE265A Winter 2019 9

 SSB NF: High-IF and Low-IF Rx

 In practice, 𝑺,𝑰𝑴 is attenuated by antenna selectivity and the tuned LNA.

 High-IF Rx
o 𝑺,𝑰𝑴 can be heavily filtered (therefore, , , )
 Low-IF Rx
o 𝑺,𝑰𝑴 is not as adequately filtered (therefore, , , )
 As a result, the “NF of the low-IF Rx” can be 3dB higher (worse) than
the “NF of the high-IF Rx”:
, 𝒉𝒊𝒈𝒉 𝑰𝑭
,
, 𝑺,𝑰𝑴
, ,
, 𝒍𝒐𝒘 𝑰𝑭
,
 Recall the trade-off between sensitivity vs. selectivity
o High-IF  better sensitivity (better attenuation of image and noise)
o Low-IF  better selectivity (easier channel select filter/ more adjacent
channel rejection)

© Leung, ECE265A Winter 2019 10

 Side-topic: passive mixer example

 Passive – when transistors are used as switches (not amplifiers)

o Switch is open/ closed at LO frequency ( )
 This is an example of a linear (with respect to the RF port) but time-
varying* circuit

* Interested students please read

Razavi text, page 10-11.
This so-called “return-to-zero” (RZ)
mixer will be further analyzed in 265B.

© Leung, ECE265A Winter 2019 11

 Side-topic: active mixer examples

 Conceptually, an active mixer can be realized by a MOS transistor’s

square-law characteristic (2nd-order nonlinearity):
o ,

* Here, we assume the transistor

stays in saturation at all time.

𝑽𝒔,𝑫𝑪

 Or more commonly, active commutating

down-conversion mixers are formed switch

by a transconductor and a pair of

“commutating switches” transconductor
 LO signals “hard-steer” tail (signal)
current to p/m output benches
© Leung, ECE265A Winter 2019 12
 Direct-conversion Receiver

 Direct-conversion (aka zero-IF, homodyne) receiver is the most popular

architecture today.
o Simpler hardware – one LO, one mixer  higher level of integration
o No image issue (need quadrature downconversion)
o Channel selection is achieved by lowpass filter
o Less mixing spurs

 But this straight-forward architecture also comes with challenges

1. LO leakage
2. DC offset
3. Even-order distortion*
4. Flicker noise
5. I/Q mismatch
6. Mixing spurs

* Next lecture
© Leung, ECE265A Winter 2019 13
 (1) LO leakage

 Strong LO signal can leak to the antenna through:

o parasitic capacitances between mixer LO-RF ports, and LNA output-
input (finite reverse isolation, ), and/ or
o substrate coupling to input
 This is problematic because (spec < -70dBm):
o It may desensitize other receivers operating in the same band
o There are straight FCC regulation (chips that violate this are unsellable)
 LO leakage also occurs in heterodyne receivers, but it can be suppressed
by the band-select filters in both the aggressor and victim receivers.
 LO leakage can be minimized by symmetric RF and LO path layout

LO leakage cancelled
by layout symmetry (diff. LO)
© Leung, ECE265A Winter 2019 14
 (2) DC offset

 DC offset can be created by LO self-mixing

o In-band LO leakage ( , ) appears at the LNA input. It is
amplified ( , ) and mixed with quadrature LO, producing different
DC baseband signals:
o ,
o ,
 This is problematic because it may saturate baseband circuits, prohibiting
signal detection. Consider the example:
10mV
o LO leakage = -60dBm (or 316uV), (= 316uV x 31.6V/V)

𝒍𝒆𝒂𝒌 𝑳𝑶 𝒍𝒆𝒂𝒌 1V (!!!)

316uV (= 10mV x 100V/V)
(LO leakage)

© Leung, ECE265A Winter 2019 15

 Side: Self-mixing in Sliding-IF Rx

 The LO1 leakage is benign

o The final self-mixing product would not be at DC
 The LO2 leakage is still a concern
o It may be less severe because the frequency is lower,
o But attention is still required for signal saturation and linearity concern

𝒇
𝟒𝒇𝒊𝒏 2nd mix: 𝒊𝒏 , or 𝒊𝒏 etc.
1st mix: 𝟑
𝟑 (Not at DC)

𝒊𝒏

LO1 leakage 𝒊𝒏
rejected by BPF
LO2 leakage
problem same as
a homodyne
© Leung, ECE265A Winter 2019 16
 DC offset mitigation (1): HPF

 Offset cancellation can be achieved by highpass filters (aka ac-coupling, or

dc-blocking)
 However, this also removes a fraction of the signal’s spectrum near DC,
introducing inter-symbol interference (ISI).
o As a rule of thumb, want f3dB < 1/1000 of the symbol rate. For example,
802.11b (20 Mbps) requires f3dB < 20 kHz, thus very big RC
o Slow settling (long time constant) will result, when Rx settings (such as
LO frequency or gain) are changed.
 Therefore, AC coupling is very rarely used.

f3dB given by
𝟏⁄ 𝟐𝝅𝑹𝟏 𝑪𝟏

© Leung, ECE265A Winter 2019 17

 DC offset mitigation (2): Feedback

 By inspection (KCL at node X), we can write:

o

o (Want for offset rejection)

→

o Pole: ; Zero:
 Unfortunately, the corner frequency ( ) is higher by
o Need even bigger capacitance for the same cutoff 
o The cap is no longer floating (but grounded)  realized by MOS-cap 

For comparison:
Approach (1): AC coupling

Vin Vout

Approach (2):
© Leung, ECE265A Winter 2019 Feedback 18
 DC offset mitigation (3): DAC

 Current (I1) produced by a DAC to cancel dc offset at X.

 Foreground calibration:
o With no Rx signals, ADC digitizes the baseband output. The proper
word (D) is applied to the DAC to minimize Vout.
o The control D is measured for all channels and Rx gain settings, and
stored in small on-chip memory (e.g., E-fuse)
 In actual Rx operation, the word D is “frozen”
o As the discrete-time feedback loop is open in operation, it would not
impact the baseband frequency response
 This approach is popular
o In modern CMOS technologies, ADC/ DAC are very small (cf. to a cap)

binary-weighted
current DAC

© Leung, ECE265A Winter 2019 19