Você está na página 1de 48

ece145c lecture notes Copyright Mark Rodwell, 2016

Mixers:
Reciprocity
N-path designs

Mark Rodwell,
University of California, Santa Barbara
Copyright Mark Rodwell, 2016

Ideal mixing is multiplication

VB (t )  VI cos(Bt ) VRF (t )  VI cos( B t )  VLO cos( LO t ) / V0


Note the coefficient 1 / V0
VLO (t )  VLO cos(LOt )

How do we actually provide multiplication ?


Copyright Mark Rodwell, 2016

Mixing: learn to do the math efficiently

Trick : 2 cos t  e jt  e  jt  z  1 / z

e j  cos( )  j sin(  )  2  cos( )  e j  e j and 2 j  sin(  )  e j  e j

2  cos( B t )  e j Bt  e  j B t  z1B  z B1


2  cos( LO t )  e j LO t  e  j LO t  z1LO  z LO
1

so
4  cos( B t )  cos( LO t )  ( z1B  z B1 )( z1LO  z LO
1
)
 z1B z1LO  z1LO z B1  z1B z LO
1
 z B1 z LO
1

 e j Bt e j LO t  e j LO t e  j Bt  e j B t e  j LO t  e  j Bt e  j LO t
 (e j B t e j LO t  e  j Bt e  j LO t )  (e j LO t e  j B t  e j B t e  j LO t )
 2  cos ( LO   B )t   2  cos ( LO   B )t 
Copyright Mark Rodwell, 2016

Mixing by time-dependent conductance


Analysis is simpler if we assume small RF and IF signals
I d  I bias exp((Vbias  Vdiode) / Vt ) where Vt  kT / q

LO voltage modulates the diode current.


If the LO is small,
I d (t )  I bias  I LO cos( LO t )

The diode conductivi ty is then


g (t )  I d (t ) / Vt  g 0  g LO cos( LO t )

The small RF and IF signals are applied to g (t ) :


Vd (t )  VRF cos( RF t )  VIF cos( IF t )
I d (t )  Vd (t ) g (t )
 g 0VRF cos( RF t )  g 0VIF cos( IF t )
 ( g LO / 2)VRF cos(( RF   LO )t ) cos(( RF   LO )t ) 
 ( g LO / 2)VIF cos(( IF   LO )t ) cos(( IF   LO )t ) 

The time - varying conductivi ty produces the sum and difference frequencie s
Copyright Mark Rodwell, 2016

The LO drive should be big !

I d (t )  g 0VRF cos( RF t )  g 0VIF cos( IF t )  Resisitive currents (loss !)


 ( g LO / 2)VRF cos(( RF   LO )t ) cos(( RF   LO )t )   mixing terms
 ( g LO / 2)VIF cos(( IF   LO )t ) cos(( IF   LO )t )   mixing terms

The local oscillator should * strongly * modulate the conductivi ty


Copyright Mark Rodwell, 2016

With big LO drive, mixer becomes a switch

G (t ) is now a square wave.


Modulation of G (t ) by LO is strong
 strong mixing signal.

G  2 cos(3 LO t ) cos(5 LO t ) 
G (t )   on   Gon  cos( LO t )    ....
 2   3 5 
 Gon 
 2   this will give us direct RF  IF coupling ...not good.
 
Gon  cos( LO t )  generates desired mixing terms ...( RF   LO )
2

2  cos(3 LO t ) 
Gon 
    generates * harmonic * mixing terms ...( RF  3 LO )
3 
Copyright Mark Rodwell, 2016

Idealized (switch) double-balanced mixer

Input is multiplied by  1,-1,1,-1,.... , i.e. by a squarewave


VIF (t )  M (t )VRF (t ) where M (t ) is a squarewave
The squarewave has a Fourier series :
4 cos(3 LO t ) cos(5 LO t ) 
M (t )  cos( t )    ....
  
LO
3 5 
So we are multiplying VRF (t ) with cos( LO t )
First hint of trouble : there' s also 3 LO , 5 LO ,...
Copyright Mark Rodwell, 2016

A Gilbert-cell mixer is an amplifier and a switch

LO drive voltage is sufficient for


upper quad to operate as switches.

Gilbert cell is :
A linear input stage
for isolation, gain, ...
...not really part of the mixer : a preamplifer
A switch, as in previous figure.
Copyright Mark Rodwell, 2016

Typical mixer presentation then moves to circuits:

You have studied such circuits before.

But, before we concentrate on circuits, how well do we understand mixing ?


Insertion loss ? Noise figure ? Image responses ? Harmonic responses ?
Input - output isolation ?
Copyright Mark Rodwell, 2016

Ideal switch-based mixer

Insertion loss ? Noise figure ? Image responses ?


Harmonic responses ? Input - output isolation ?

We can learn much by studying an ideal mixer


Copyright Mark Rodwell, 2016

What is our goal ?

Question : Why do we use LNAs in receivers ?


Answer : to reduce the mixer' s noise figure contribution.
Question : Why do mixers have 3 - 6 dB noise figures ?

If low mixer noise figure  eliminate LNA.


Lower cost, higher receiver IP3.
Note : IF amp operates at lower frequency, can have low Fmin
Copyright Mark Rodwell, 2016

Even ideal mixers have:


Image response  out - of - band response, added noise.
LO harmonic response  out - of - band response, added noise.
Attenuation : becuase input signal is converted to several frequencies
Bilateral response : output couples back to input. At several frequencies.

Ideal mixer

Ideal mixer with filters.


This behaves differently
from the mixer
with no filter.
Copyright Mark Rodwell, 2016

Image response→ interference, loss of SNR


VIF (t )  M (t )VRF (t )
4 cos(3 LO t ) cos(5 LO t ) 
M (t )  cos( t )    ....
  
LO
3 5 

Image response :
If f IF  f RF  f LO , then f image  f LO  f IF  f RF  2 f LO
Signals and noise at f image also mix to f IF .

Problem #1 is interference :
RF front - end needs filter to reject f image

Problem #2 is loss in SNR due to f image


Mixer input noise power spectral density @ f RF  kTFGLNA ( f RF )
Mixer input noise power spectral density @ f image  kTFGLNA ( f image )
Poor front - end filtering ?  Image response adds significant noise
Copyright Mark Rodwell, 2016

But there is *a different* image response problem:


VIF (t )  M (t )VRF (t )
4 cos(3 LO t ) cos(5 LO t ) 
M (t )  cos( t )    ....
  
LO
3 5 

Image response at RF : signals and noise at f image also mix to f IF .


If f IF  f RF  f LO , then f image  f LO  f IF  f RF  2 f LO

But, there is another problem :


If the input is at f RF , then one output is at f IF  f RF  f LO
If the input is at f RF , another output is at f IF '  f RF  f LO
If the mixer is passive and lossless, it conserves energy !
Signal power at f IF '  less power at f IF   attenuation.
 less signal power power at IF amp
 IF amp noise has greater effect on receiver SNR
Copyright Mark Rodwell, 2016

LO harmonics also produce images:


VIF (t )  M (t )VRF (t )
4 cos(3 LO t ) cos(5 LO t ) 
M (t )  cos( t )    ....
  
LO
3 5 

At the RF port, other frequencies also mix to f IF


3 f LO  f IF , 5 f LO  f IF , 7 f LO  f IF ...
And, given imperfect LO symmetry : 2 f LO  f IF , 4 f LO  f IF , ...
Problem : out - of - band interference  need good RF filter
Problem : out - of - band noise contribution  need good RF filter

Further, an input at f RF mixes to many output frequencies


f RF  f LO , f RF  2 f LO , f RF  3 f LO , ...
signal power at these  less power at f IF  more attenuation
Copyright Mark Rodwell, 2016

Ideally: eliminate spurious responses with filters


VIF (t )  M (t )VRF (t )
4 cos(3 LO t ) cos(5 LO t ) 
M (t )  cos( t )    ....
  
LO
3 5 

But :
off - wafer filters cost money, increase product size
on - wafer filters occupy die area, & are low - Q
If f IF / f RF is small, then the filter must be very high Q

* And* :
In a real mixer, the switches are diodes or transistors
these have RC parasitics, shot noise generators.
1) the resistors & transistor junctions will introduce kTF , shot noise directly at f RF , f IF
2) the resistors also generate noise at all the image frequencies.
These will also mix into the receiver passband;
and external filters cannot prevent this.
Copyright Mark Rodwell, 2016

Real mixers are bilateral, not unilateral


VRF (t )  M (t )VIF (t )
4 cos(3 LO t ) cos(5 LO t ) 
M (t )  cos( t )    ....
  
LO
3 5 

Apply signal at f IF to the IF port

RF port :
response @ f LO  f IF  f RF
response @ f LO  f IF
response @ 2 f LO  f IF
response @ 2 f LO  f IF , etc.

Just like other circuits, mixers are bilateral ( S 21S12  0)


Passive mixers S 21  S12
Active mixers S12  S 21 but S12 nevertheless  0.

Note : Sij here defined at different frequencies for input & output ports.
Copyright Mark Rodwell, 2016

Diode Double Balanced Mixer: Two-Port Representation

If we apply filters, as shown, to restrict t he signal


frequencie s at the two ports, we can again represent
the mixer with a 2 - port network, where the two ports
have signals at frequencie s  RF and  IF .

 Mixers have MAG, optimum impedances, etc.

Derivation is not hard. But, we will not pursue here.


For ideal switches, Y , Z matrices will have infinities.
in that case, use S matrix.
Copyright Mark Rodwell, 2016

Bilateral mixing→ spurious RF responses


Example : filter at IF port, not at RF port
Apply signal at f RF to the IF port
 produces signal at f IF at the IF port
 then produces signal at the RF port
@ f LO  f IF  f RF
@ f LO  f IF
@ 2 f LO  f IF
@ 2 f LO  f IF , etc.
Out - of - band signal responses.
Antenna will re - radiate.
Suppressed by LNA S12 , if present.
Suppressed by filter, if present,
and if filter is sufficiently narrow.
(one response is @ f RF  2 f IF )

This is in addition to LO leakage : also radiates from antenna; much stronger signal
Copyright Mark Rodwell, 2016

Eliminating Noise from Image Response

Image - reject mixer suppresses both


image signal and image noise response

Trap provides zero available noise


power at image frequency

Filtering : ~ kT noise at image frequency


but ~ kTFG noise at signal frequency
Copyright Mark Rodwell, 2016

System-Level Mixer Noise analysis


Citation: TUTORIAL 5594: System Noise-Figure Analysis for Modern Radio Receivers By: Charles Razzell,Maxim Integrated Products, Inc.

Model of Mixer' s internal noise.


Adds N s at f signal ; spectral density S s
Adds N i at f image ; spectral density S I
Adds N IF at f IF ; spectral density S IF
GS  mixer gain for signal frequency
GI  mixer gain for image frequency

Total noise at IF port


S mixer  S S GS  S I GI  S IF
Again, this is just the mixer' s internal noise.
Copyright Mark Rodwell, 2016

System-Level Mixer Noise analysis


Receiver model :
LNA adds noise N LNA, S at f signal
LNA has gain GLNA, S at f signal
LNA adds noise N LNA, I at f image
LNA has gain GLNA, I at f image
Include filter 2 frequency reponse in GLNA

IF signal power
PIF  Psignal GLNA, S GS
IF noise power spectral density
S IF  (kT  N LNA, S )GLNA, S GS  (kT  N LNA, I )GLNA, I GIM  S mixer
 kTFLNA, S GLNA, S GS  kTFLNA, I GLNA, I GIM  S mixer
Component of IF noise power spectral density from RF source @f S
S IF ,from antenna@f s  kTGLNA, S GS
System Noise figure
FLNA, S GLNA, S GS  FLNA, I GLNA, I GIM  S mixer / kT
Fsystem  S IF / S IF ,from antenna@f s 
GLNA, S GS
GLNA, I GIM S mixer
Fsystem  FLNA, S  FLNA, I 
GLNA, S GS kTGLNA, S GS
Copyright Mark Rodwell, 2016

System-Level Mixer Noise analysis


System Noise figure
GLNA, I GIM S mixer
Fsystem  FLNA, S  FLNA, I 
GLNA, S GS kTGLNA, S GS

Suppose : no RF filtering of image,


GLNA, I  GLNA, S , GIM  GS
S mixer
Fsystem  FLNA, S  FLNA, I 
kTGLNA, S GS
we have doubled the LNA noise contribution

Suppose : perfect RF filtering of image,


GLNA, I  0
S mixer
Fsystem  FLNA, S 
kTGLNA, S GS
we have elimnated the image noise contribution
except for that internal to the mixer
Copyright Mark Rodwell, 2016

System-Level Mixer Noise analysis: another case


Assume :
filter 2 provides perfect RF filtering of image,
but we then have another gain stage (LNA 2).
GLNA1, I  0
FLNA 2, S  1 FLNA 2, I  1 GLNA 2, I GIM S mixer
Fsystem  FLNA1, S   
GLNA1, S GLNA1, S GLNA 2, S GS kTGLNA1, S GLNA 2, S GS

We have doubled the noise contribution of LNA 2.


Copyright Mark Rodwell, 2016

System-Level Mixer Noise analysis: Big Picture


Receiver noise model
If unfiltered, image responses will
add LNA (etc) noise at image frequency
to that at signal frequency
 increased receiver noise

Mixer noise model


Mixer internal noise at image frequency
adds to that at signal frequency
 increased mixer noise
This can' t be filtered.

Image responses also arise from LO harmonics

Clearly : try to minimize mixer harmonic and image responses


Copyright Mark Rodwell, 2016

Eliminating Noise from Image Response

Image - reject mixer suppresses both


image signal and image noise response

Trap provides zero available noise


power at image frequency

Filtering : ~ kT noise at image frequency


but ~ kTFG noise at signal frequency
Copyright Mark Rodwell, 2016

N-path mixers: Motivation

Why use LNAs in superheterodyne receivers ?


...because mixer has high noise figure.

LNAs degrade overall receiver overload/jammer tolerance:


-Receiver IIP3 set by LNA IIP3 & (mixer IIP3)/(LNA gain).
-Fixed-tuned RF filter can't protect LNA & mixer
from interferers within receiver overall tuning bandwidth.

If we could make a very low noise figure mixer….


LNA would not be needed--> improved receiver IP3.
Out of band IM3 from mixer alone.
Low receiver noise (IF amp can have low Fmin).
Copyright Mark Rodwell, 2016

N-path mixers: Motivation

Why do mixers have high noise figure ?


noise partly from component noise & attenuation.
noise partly from mixer image & harmonic responses.

Can we make mixers with very low noise figure ?


good devices→ low component noise.
subsampling method→ low/zero image responses.
Copyright Mark Rodwell, 2016

Where are the harmonic responses coming from ?

We are mixing against a squarewave.


...not a sinewave.

Yet, we wish to mix using switches, not


small - signal nonlinear elements

How might we do this ?


Copyright Mark Rodwell, 2016

Array of sample-hold gates:

This takes time - samples of the input waveform.


Several samples per signal cycle.
Samples taken with switches.
Switch mixer, not V 2 /V0 mixer.

There are N switches


Number them 1,..., n,..., N
Copyright Mark Rodwell, 2016

Array of sample-hold gates: now sum the outputs

Form linear sums of the S/H outputs.


weightings : cos( )  cos(2n / N )
Approximates multiplying Vin with cos(( LO / N )t )
Yet, we are using switches, not x 2 nonlinearities
Copyright Mark Rodwell, 2016

Quadrature mixer with sample-hold array

Provide both cosine and sine weightings :


cos( )  cos(2n / N ), sin ( )  sin( 2n / N )
 quadrature mixer.
Copyright Mark Rodwell, 2016

Array of sample-hold gates: harmonic tuning

The desired harmonic can be selected


Copyright Mark Rodwell, 2016

Selecting Harmonics:
One S/H : mixing
against pulse train
 many LO
harmonics

Selecting
f LO / N

Selecting
2 f LO / N
Copyright Mark Rodwell, 2016

N-path Filter Has No Image Responses

Array has N non-overlapping switches


unfiltered outputs: positive & negative IF images
unfiltered outputs: responses for all LO harmonics
sine/cosine weighting: selects particular LO response
sine / cosine quadrature summing suppresses IF image
Input RF filter suppresses responses beyond N*fLO.
Copyright Mark Rodwell, 2016

What is the bandwidth ?

Assume :
f signal and f LO / N are  baseband signal frequency

 use switched - capacitor theory :


Switch duty cycle  1 / N
Equivalent impedance  NZ 0

 Bandwidth  1/2 ( NZ 0 )C
Copyright Mark Rodwell, 2016

N-path filter tuning


The receiver signal frequency is tuned by adjusting:
The LO (sampling) frequency (fsig = 2 fLO + fIF )
The selected LO harmonic (cos/sin weighting)
Wide tuning range is easily achieved
Copyright Mark Rodwell, 2016

N-path filter with a nonzero IF Frequency

Input gates are sample-and-filter, not sample-and-hold


No in-band resonant enhancement of voltage or current
Out-of-band voltage is suppressed at switch
→ System IM3 set by switch stage, no Q:1 degradation in dynamic range
As N → ∞ perfect conversion, image response → 0, NF →0
Copyright Mark Rodwell, 2016

With nonzero IF output frequency

SSB mixing is then obtained by summing


the I and Q outputs at 900 relative phase
Copyright Mark Rodwell, 2016

Again, what is the bandwidth ?

Assume :
f signal and f LO / N are  baseband signal frequency

 use switched - capacitor theory :


Switch duty cycle  1 / N
Equivalent impedance  NZ 0

 Bandwidth  1/2 ( NZ 0 )C
Copyright Mark Rodwell, 2016

Simulations: Late 2008, early 2009

Simulation example showing the 8 sampling phases, and the sine and cosine weighted outputs both before and
after 90 degree phase-shifting
Copyright Mark Rodwell, 2016

Simulation

CMOS switch array responses with 1st and 3rd LO harmonic tuning.
The IF bandwidth is set at 30 MHz equivalent to a Q of 100
Copyright Mark Rodwell, 2016

Simulations
device model: 35 nm InP/InGaAs HEMT, 400-500 GHz ft , fmaxf compared to 65nm NMOS (state of art at time of simulation)

3
Increased N : File:nf-vs-N-ssampRLC

65nm CMOS (digital)


better approximation to sinewave 35nm Std InP HEMT
35nm Low Noise InP HEMT

Noise Figure (dB)


reduced LO harmonic content 2

 reduced noise figure.


1

ADS Sim mark4

0
2 4 6 8 10 12 14 16
Number of Samplers
Copyright Mark Rodwell, 2016

Dynamic Range

Input Capacitor C is large i.e. Z0C >> TON


Within passband, C charges to RF signal,
Current→ 0, distortion is minimized, IF amp determines IP3
Outside passband, C discharged
Current ~ Isig distortion is from current in switch element
No resonant enhancement in either case
No 3dB noise from image/harmonic response in limit as n→∞
No RF filtering needed since no LNA and mixer has high IIP3
Copyright Mark Rodwell, 2016

Simulations
device model: 35 nm InP/InGaAs HEMT, 400-500 GHz ft , fmaxf
compared to 65nm NMOS (state of art at time of simulation)
Copyright Mark Rodwell, 2016

65nm MOSFET Simulations – 8 element array

Simulation of in-band IM3, showing 12 dBm IIP3.


Simulations showed 1.7 dB NF at 1.2 GHz
12.5 dBm IIP3 is obtained with mixing against the
2nd LO harmonic (2.2 GHz inputs).

Out-of-band IM3 simulation


~0 dBm equivalent IIP3.
Signals selected so that the IM3 response in-
band but carriers are not.
IM3 products are 20 dB stronger, so the
"equivalent" IP3 has dropped 10 dB
Copyright Mark Rodwell, 2016

65nm MOSFET Simulations – 8 element array

• Simulation of in-band IM3, showing 12


dBm IIP3.
• Using the 65 nm MOS model, simulations
showed 1.7 dB NF at 1.2 GHz
• 12.5 dBm IIP3 is obtained with mixing
against the 2nd LO harmonic (2.2 GHz
inputs).

• Out-of-band IM3 simulation indicating ~0


dBm equivalent IIP3.
• Signals selected so that the IM3 response
in-band but carriers are not.
• IM3 products are 20 dB stronger, so the
"equivalent" IP3 has dropped 10 dB
Copyright Mark Rodwell, 2016

N-path: first demonstrated early 2009

Você também pode gostar