Professional Documents
Culture Documents
Copyright
Fei
c Yuan 2016
Preface
1
Contents
1 Wire Channels 5
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Laboratory Work . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1 PRBS Generator . . . . . . . . . . . . . . . . . . . . . 6
2.2 Current-Mode Driver . . . . . . . . . . . . . . . . . . . 6
2.3 Wire Channel . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3 Laboratory Report . . . . . . . . . . . . . . . . . . . . . . . . 8
3 Phase-Locked Loops 16
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.1 Voltage-Controlled Ring Oscillators . . . . . . . . . . . 16
1.2 Bang-Bang Phase Detector . . . . . . . . . . . . . . . . 19
1.3 Phase-Locked Loop . . . . . . . . . . . . . . . . . . . . 22
2 Laboratory Work . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.1 Charge Pumps . . . . . . . . . . . . . . . . . . . . . . 23
2
2.2 Loop Filters . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Open-Loop Test of PLL . . . . . . . . . . . . . . . . . 24
2.4 Closed-Loop Test of PLL . . . . . . . . . . . . . . . . . 25
3 Laboratory Report . . . . . . . . . . . . . . . . . . . . . . . . 26
4 Pre-Emphasis 29
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.1 Unit Delay Cells . . . . . . . . . . . . . . . . . . . . . 30
1.2 Pre-Emphasis Tap Multiplier . . . . . . . . . . . . . . 30
1.3 Summer/Driver . . . . . . . . . . . . . . . . . . . . . . 31
2 Laboratory Work . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1 Serial Link . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Eye Diagram . . . . . . . . . . . . . . . . . . . . . . . 32
2.3 Power Consumption . . . . . . . . . . . . . . . . . . . 32
3 Laboratory Report . . . . . . . . . . . . . . . . . . . . . . . . 33
3
4
Wire Channels
1 Introduction
Wire channels over which data are transmitted include printed-circuit-board
(PCB) traces, interconnects, vias, connectors, etc. Due to resistive, capaci-
tive and inductive losses at high frequencies1 , wire channels suffer from finite
bandwidth, which is the main cause of inter-symbol interference (ISI). In this
laboratory, you will study the characteristics of wire channels in both time
and frequency domains. You will design a 1 Gbps serial link consisting of a
pseudo-random bit sequence (PRBS) generator, a current-mode transmitter,
a wire channel modeled as a transmission line, and termination resistors.
1
Resistive loss also known as ohmic loss is due to skin effect and proximity effect of
conductors at high frequencies. Capacitive loss also known as dielectric loss arises from
the shunt capacitance between channels and ground planes/substrates. Inductive loss is
caused by the inductance of channels.
5
CHAPTER 1. WIRE CHANNELS 6
2 Laboratory Work
2.1 PRBS Generator
The PRBS generator used in this laboratory is a behavioral functional block
coded in Verilog. It outputs a pseudo-random bit-sequence. It is located in
the library called ahdlLib, cell rand-bit-stream.
2.4 Analysis
The following analyses are to be performed :
3 Laboratory Report
Laboratory reports must be prepared in either Microsoft Word, LaTeX, or
other text editors. Figures, tables, and expressions should be embedded
in the main body of reports. All expressions, figures, and tables must be
numbered. All figures and tables must be captioned. Cadence print-outs
attached to the end of laboratory reports are not acceptable. Hand-
written laboratory reports will be rejected. Laboratory reports must
contain the followings :
3. Conclusions (10%)
Provide no less than three (3) but no more than five (5) conclusions
that you can draw from this design laboratory. Your conclusions must
be supported by the simulation results obtained and must be as specific
and precise as possible.
Chapter 2
Continuous-Time Linear
Equalizers
1 Introduction
It was observed in Laboratory 1 that the longer the channel, the severer the
attenuation of the signal traveling along the channel. One effective way to
compensate for the loss of the channel is to boost the received signal at the
receiving end using an amplifier. In order to cope with the timing constraint
of multiple Gbps data rates, the amplifier must have a large bandwidth.
Since the effect of the finite bandwidth of the channel is the attenuation
of the high-frequency components of the transmitted signal (Note that the
transmitted signal is broadband and contains both low and high-frequency
components) while having no effect on the low-frequency components of the
transmitted signal, the amplifier should therefore behave in such a way that
it provides a small voltage gain at low frequencies and a large voltage gain
at high frequencies, typically at the baud-rate frequency of the transmitted
signal. Although it seems that the larger the voltage gain of the amplifier,
the better the compensation, a large voltage gain at high frequencies will
also deteriorate crosstalk, which typically intensifies at high frequencies. At
a result, only a moderate gain at high frequencies should be provided by the
10
CHAPTER 2. CONTINUOUS-TIME LINEAR EQUALIZERS 11
2 Laboratory Work
2.1 Serial Link with CTLE
Connect CTLE of Fig.2.1 to the output of the serial link designed in Labo-
ratory 1, as shown in Fig.2.2.
One important issue that needs to be addressed is the common-mode
input voltage of CTLE. The 50Ω termination resistors at the far end of
the channel will result in a large common-mode input voltage of CTLE,
which might give rise to an improper dc bias of the NMOS transistors of
CTLE subsequently a low voltage gain. The tail current sources and load
resistors of CTLE should be carefully chosen such that the NMOS transistors
are properly biased in saturation in order to have a large transconductance
subsequently a large voltage gain. In order to have a large voltage gain
at high frequencies, the load resistors of CTLE can be replaced with spiral
inductors. The added inductors resonate out the load capacitance of the
CTLE at the desired frequency, such as the baud-rate frequency of the serial
link so as to boost the bandwidth.
Another important issue is the choice of the value of Rf and Cf . Using
small-signal analysis one can show that the CTLE has a zero at frequency 1
1
ωz = (2.1)
Rf Cf
and a pole at frequency
1 R f gm Rf gm
ωp = + 1 = ωz +1 . (2.2)
Rf Cf 2 2
Since ωz < ωp , the zero boosts the gain of CTLE at 20 dB/dec when
ω > ωz . If the channel is modeled as a single-pole system with bandwidth
ωch , then ωz should be set in the vicinity of ωch so as to effectively compensate
for the loss of the channel.
1
These results are obtained under the assumption that Cgs , Cgd Cf and RL ro .
CHAPTER 2. CONTINUOUS-TIME LINEAR EQUALIZERS 13
The resistance of the load resistors should also be chosen with care. A
large load resistance will certainly benefit voltage gain. This, however, is at
the expense of reduced bandwidth as the output pole is formed by the load
resistor and the capacitance at the output of CTLE. A large load resistance
will also lower the common-mode voltage of the output of CTLE. Note that
in a typical DFE, CTLE is followed by a slicer (comparator). The common-
mode output voltage of CLTE affects the operation of the slicer and needs
to be set as per the common-mode input voltage range of the slicer.
2.2 Analysis
The following analyses are to be performed :
period 20 ns. Plot the time-domain response vo (t) of CTLE for three
difference lengths of the channel (10 cm, 50 cm, and 100 cm). Measure
the pre-cursors, main cursor, and post-cursors of vo (t). Compare these
results with those without the CTLE obtained in Laboratory 1.
3 Laboratory Report
Laboratory reports must be prepared in either Microsoft Word, LaTeX, or
other text editors. Figures, tables, and expressions should be embedded
in the main body of reports. All expressions, figures, and tables must be
numbered. All figures and tables must be captioned. Cadence print-outs
attached to the end of laboratory reports are not acceptable. Hand-
written laboratory reports will be rejected. Laboratory reports must
contain the followings :
3. Conclusions (20%)
Provide no less than three (3) but no more than five (5) conclusions that
you can draw from this design laboratory with a particular focus on the
comparison of the results obtained without CTLE in Laboratory 1 and
those obtained in this laboratory. Your conclusions must be supported
by the simulation results obtained and must be as specific and precise
as possible.
Chapter 3
Phase-Locked Loops
1 Introduction
Phase locked loops are used extensively in data communications, computer
systems, RF communications, instrumentation, and signal processing. In se-
rial links, PLLs are used at both transmitters and receivers. In transmitters,
PLLs are locked to a clean reference, often the output of a crystal oscilla-
tor. The clocks generated by the PLLs are used to perform important tasks
such as pre-emphasis. In receivers, PLLs are locked to incoming data sym-
bols with embedded timing information to perform clock recovery. In this
laboratory, you are required to design a PLL with an Alexander bang-bang
phase detector, a 4-stage voltage-controlled ring oscillator, and a 500 MHz
50% duty-cycle digital input.
16
CHAPTER 3. PHASE-LOCKED LOOPS 17
tive latch to sharpen the falling edge of the waveform of the oscillator1 . As
pointed out by Weigandt et al. that the noise of VCOs at threshold-crossing
shifts threshold-crossing time by an amount that is proportional to the noise
injected at the threshold-crossing and inversely proportional to the slew rate
of the output voltage2 :
vn2
∆τ 2 = , (3.1)
(dvo /dt)2
where ∆τ 2 and vn2 are the power of timing jitter and that of the noise injected
at the threshold-crossing, respectively and dvo /dt is the slew rate of the
output voltage of the oscillator at the threshold-crossing. It is seen from (3.1)
that sharpening transition edges lowering timing jitter. Reducing the
timing jitter or phase noise of VCOs is critical to lower the overall phase
noise of PLLs. As the output of the VCO is used to sample incoming data,
minimizing the timing jitter of the sampling clock, i.e., the output of the
VCO, is pivotal to the reliable recovery of data especially when data rate is
high.
Laboratory Work
2. Rise and Fall times : Perform the time-domain analysis of the oscillator.
The oscillator can be activated by connecting a small capacitor with
1
X. Mailand, F. Devisch, and M. Kuijk, “A 900 Mb/s CMOS data recovery DLL using
half-frequency clock,” IEEE Journal of Solid-State Circuits, vol. 37, No. 6, pp 711-715,
June 2002.
2
T. Weigandt, B. Kim, and P. Grey, “Analysis of timing jitter in ring oscillators,” Proc.
of IEEE International Symp. Circuits and Systems, pp. 27-30, London, 1994.
CHAPTER 3. PHASE-LOCKED LOOPS 18
an initial voltage to the output node of one of the delay stages of the
oscillator. Note that the value of the capacitor should be small in order
to minimize its loading effect. Alternatively, you can inject a current
pulse of a narrow pulse width and a long period into any node of the
oscillator to start oscillation. Both methods mimic the injection of
noise to start oscillation in real oscillators. Plot the output voltage
of the each stage of the oscillator. Measure the rise and fall times of
the output voltage of the oscillator. Note that transistors should be
properly size in order to have the same rise and fall times3 .
fmax − fmin
Kvco = (3.2)
Vc,max − Vc,min
over the linear portion of the frequency tuning curve where fmax and
fmin , Vc,max and Vc,min are the maximum and minimum frequencies
and their corresponding control voltages of the oscillator in the linear
portion of the frequency tuning curve, respectively. Note the unit of
Kvco is Hz/V.
Since a large number of DFFs are used in Alexander PDs, the character-
istics of DFFs undoubtedly affects the performance of the PDs. There are
many ways to construct a DFF. The conventional DFF shown in Fig.3.4(a)
suffers from a large transistor count, high power consumption, and a long
propagation delay. True-single-phase-clocking (TSPC) logic circuits pro-
posed by Yuan and Svensson7 offer the advantages of a low transistor count,
5
J. Alexander, “Clock recovery from random binary signals,” IEE Electronics Letters,
Vol. 11, No. 22, pp. 541-542, 1975.
6
B. Razavi, “Challenges in the design of high-speed clock and data recovery circuits,”
IEEE Comm. Mag., pp. 94-101, Aug. 2002.
7
J. Yuan and C. Svensson, “New single-clock CMOS latches and flipflops with improved
CHAPTER 3. PHASE-LOCKED LOOPS 21
a small chip area, low power consumption, and a fast transient response. In
this laboratory, TSPC positive-edge triggered DFF shown in Fig.3.4(b) is
used to construct Alexandar PDs8 .
Laboratory Work
Apply two 500 MHz digital signals of different phases to the input of the
phase detector. Vary the phase of the two input signals in such a way that
one leads the other or one lags the other and record the output of the phase
detector. Plot the output of the phase detector versus the phase difference
between the two inputs and determine the minimum phase difference that
the phase detector cannot detect.
speed and power savings,” IEEE J. Solid-State Circuits, Vol. 32, No.1, pp. 62-69, Jan.
1997.
8
For various configurations of TSPC dynamic circuits, please see text book : J. Rabaey,
Digital Integrated Circuits : A Design Perspective, Upper Saddle River, NJ : Prentice-Hall,
1996. The 2nd edition of the book is also available.
CHAPTER 3. PHASE-LOCKED LOOPS 22
Figure 3.4: (a) DFF using standard logic cells. (b) TSPC positive-edge
triggered DFF.
2 Laboratory Work
2.1 Charge Pumps
The main functionality of a charge pump is to map UP and DN from the
preceding PD to a current of constant amplitude but different directions.
Fig.3.6 is the schematic of a current-steering charge pump. The charge pump
conveys a constant current JU P to or sinks a constant current JDN from the
output node, as shown in Fig.3.6. For a rich collection of charge pumps,
please see text CMOS Current-Mode Circuits for Data Communications (F.
Yuan, Springer, 2008).
• TSPC DFFs : Unlike DFFs implemented using static logic gates, TSPC
logic circuits are dynamic circuits whose logic states are retained by the
charge of the capacitors between neighboring stages. These capacitors
are formed by the input capacitance of the driven stage and the output
capacitance of the driving stage. As the voltages of these capacitors
might not be full swing (0-1.2V), a care must be taken in sizing tran-
sistors to ensure that the voltage swing of these capacitors exceeds the
threshold voltage of the driven stage so as to ensure the proper and
reliable operation of the DFFs.
10
Note that the output of the local oscillator of PLLs typically does not have 0-1.2V
swing. An voltage buffer such as a inverter chain connected to the output of the oscillator
is often used.
CHAPTER 3. PHASE-LOCKED LOOPS 26
• The current of the charge pump needs to be chosen with a care. If the
current is overly small, the resultant voltage change of the loop filter
will be too small to provide an adequate adjustment of the frequency
of the oscillator. As a result, the PLL might not be able to achieve
lock. If the current of the charge pump is overlay large, the resultant
voltage change of the loop filter will be too large to overly adjust the
frequency of the oscillator, which will also not be able to achieve lock.
3 Laboratory Report
Laboratory reports must be prepared in either Microsoft Word, LaTeX, or
other text editors. Figures, tables, and expressions should be embedded
in the main body of reports. All expressions, figures, and tables must be
numbered. All figures and tables must be captioned. Cadence print-outs
attached to the end of laboratory reports are not acceptable. Hand-
written laboratory reports will be rejected. Laboratory reports must
contain the followings :
CHAPTER 3. PHASE-LOCKED LOOPS 27
3. Conclusions (20%)
Provide no less than three (3) but no more than five (5) conclusions
that you can draw from this design laboratory. Your conclusions must
be supported by the simulation results obtained and must be as specific
and precise as possible.
11
The lock time of a PLL is the time from the injection of the input of the PLL to the
time instance at which the control voltage of the oscillator converges.
Chapter 4
Pre-Emphasis
1 Introduction
Pre-emphasis is an effective channel equalization technique performed on the
transmitter side prior to data transmission. It purposely pre-distorts data
symbols to be transmitted at the near end of the channel prior to their trans-
mission in such a way that when the pre-distorted data symbols reach the
far end of the channel, they will emerge as desired data symbols, i.e., data
symbols that yield the optimal bit error rate. Since the main impact of chan-
nels is to attenuate the high-frequency components of data symbols, arising
from the finite bandwidth of the channels, pre-emphasis can be performed
in one of the following two different ways to achieve the same results : (i)
Amplify the high-frequency components of data symbols or (ii) attenuate the
low-frequency components of data symbols. The former is undesirable as it
signifies crosstalk, which typically identifies at high frequencies. Although
the latter deteriorates signal-to-noise ratio as most of the energy of data
symbols is located at frequency of the baud rate of data symbols, it is still a
preferred choice as it does not deteriorate crosstalk.
In this laboratory, you are required to design a 2-tap pre-emphasis block
that equalizes the channel studied in previous laboratories. The pre-emphasis
block consists of (i) two unit interval (UI) delay blocks whose propagation
29
CHAPTER 4. PRE-EMPHASIS 30
delay is the symbol time, (ii) two tap multipliers that multiply the delayed
version of the data to be transmitted with tap coefficients, (iii) one summer
that constructs the pre-emphasized data symbols, and (vi) a phase-locked
loop that provides clocking signals for delay units.
1.3 Summer/Driver
The summation of the products of the delayed version of the data to be
transmitted and their corresponding pre-emphasis tap coefficients is typically
performed in the current domain so as to take the advantages of the speed
advantages of current-mode circuits. When the number of pre-emphasis taps
is large, the capacitance encountered at the summing node becomes large.
This will have a detrimental effect on the speed of the transmitter. In order
to lower the constant of the summing node, pre-emphasis block and driver
are often combined such that although the capacitance of the summing node
is large, the resistance seen at the summing node is only 50Ω, effectively
minimizing the time constant of the summing node, as shown in Fig.4.3.
2 Laboratory Work
2.1 Serial Link
Construct the serial link consisting of the PRBS generator, pre-emphasis
block, and channel designed previously as per Fig.4.4.
CHAPTER 4. PRE-EMPHASIS 32
3 Laboratory Report
Laboratory reports must be prepared in either Microsoft Word, LaTeX, or
other text editors. Figures, tables, and expressions should be embedded
in the main body of reports. All expressions, figures, and tables must be
numbered. All figures and tables must be captioned. Cadence print-outs
attached to the end of laboratory reports are not acceptable. Hand-
written laboratory reports will be rejected. Laboratory reports must
contain the followings :
• Schematics (20%)
– The schematic of the pre-emphasis block including the driver.
– The complete schematic (block-level diagram) of the serial
link.
• The eye diagram of the serial link without pre-emphasis block
(20%).
• The eye diagram of the serial link with pre-emphasis block (20%).,
• The plot of the current drawn by the pre-emphasis block and PLL
and a table tabulating the power consumption of the pre-emphasis
block and PLL (20%).
CHAPTER 4. PRE-EMPHASIS 34
3. Conclusions (10%)
Provide no less than three (3) but no more than five (5) conclusions
that you can draw from this design laboratory. Your conclusions must
be supported by the simulation results obtained and must be as specific
and precise as possible.
Chapter 5
1 Introduction
Decision feedback equalization is the most widely used post channel equal-
ization technique for multi-Gbps serial links. In this laboratory, you are
required to design a 2-tap decision feedback equalizer that will utilize blocks
such as the transmitter, channel, continuous-time linear equalizer, and PLL
designed in the previous laboratories. A 2-tap decision feedback equalizer
consists of (i) a slicer that samples the difference between the output of the
continuous-time linear equalizer and the sum of the DFE taps, (ii) 2 unit
interval (UI) delay blocks whose propagation delay is the symbol time, (iii)
2 tap multipliers, each multiplies the past decision of the slicer with a tap
coefficient, and (iv) a summer that subtracts the DFE taps from the output
of the continuous-time linear equalizer.
1.1 Slicer
Slicers are essentially voltage comparators that compare the difference be-
tween the output of the continuous-time linear equalizer and the DFE taps
with a threshold reference and produce a digital output. The threshold ref-
erence is typically the common-mode voltage of the difference signal. Fig.5.1
35
CHAPTER 5. DECISION FEEDBACK EQUALIZERS 36
Figure 5.3: Tap multiplier. The past decision is the current-steering signal,
the tap coefficient is the tail current, and the product of the past decision
and the tap coefficient is the steered current.
CHAPTER 5. DECISION FEEDBACK EQUALIZERS 38
1.4 Summer
The subtraction of DFE taps from the output of the continuous-time linear
equalizer is performed in the current domain to take the speed advantage
of current-mode circuits, as shown in Fig.5.4. The polarity of the current-
steering signals D1 , D1 , D2 , and D2 should be arranged in such a way that
the currents steered by them are subtracted from the output current of the
continuous-time linear equalizer. Since the time constant of the output of the
summer is the product of the output capacitance of the summer and the load
resistance of the CTLE, and the output capacitance rises with the increase in
the number of DFE taps, the load resistance of CTLE should be kept small.
If the load resistance is overly low, the output voltage of the summer will be
too small to be picked up reliably by the downstream slicer. To overcome
this difficulty, inductors are often added in series with the load resistors to
boost the gain of the summer at high frequencies without using large load
resistors.
2 Laboratory Work
2.1 Serial Link
Construct a 1 Gbps serial link that consists of the PRBS generator, pre-
emphasis block, channel, continuous-time linear equalizer, and PLL at the
transmitter and PLL at the receiver designed previously, as well as the DFE
detailed earlier as per Fig.5.5. For the pre-emphasis block, use the coefficients
obtained in previous laboratory. Optimal DFE tap coefficients are to be
obtained in this laboratory using a trial-and-error approach.
Figure 5.5: Serial link with a 2-tap pre-emphasis block and a 2-tap decision
feedback equalizer.
3 Laboratory Report
Laboratory reports must be prepared in either Microsoft Word, LaTeX, or
other text editors. Figures, tables, and expressions should be embedded
in the main body of the report. All expressions, figures, and tables must
be numbered. All figures and tables must be captioned. Cadence print-
outs attached to the end of the report are not acceptable. Hand-
written post-laboratory reports will be rejected. Laboratory reports
must contain the followings :
• Schematics (20%)
– The schematic of the slicer
– The schematic of the delay unit
– The schematic of tap multiplier
– The complete schematic (block-level diagram) of the serial
link.
• The eye diagram of the serial link without DFE (20%).
• The eye diagram of the serial link with DFE (20%).,
• The plots of the current drawn by (i) the pre-emphasis block, (ii)
transmitter PLL, (iii) CTLE, (iv) receiver PLL, and (v) DFE and
a table tabulating the average power consumption of these blocks
(20%).
3. Conclusions (10%)
Provide no less than three (3) but no more than five (5) conclusions
that you can draw from this design laboratory. Your conclusions must
be supported by the simulation results obtained and must be as specific
and precise as possible.
Chapter 6
1 Introduction
As the timing information of serial links is embedded in data symbols using
data encoding such as 8/10 encoding, clock recovery operation recovers the
timing information of serial links from received data symbols by detecting the
periodic transitions of data symbols. There are two widely used approaches
for clock recovery, namely phase-tracking and phase-picking. The former
utilizes a phase-locked loop at the receiver end to align the output of the
oscillator of the PLL with the transition edge of received data while the
latter allocates the transition edge of received data by sampling each data
symbol multiple times. Phase-tracking is a closed-loop approach and enjoys
the advantages of a low sensitivity to the effect of PVT (process, voltage,
and temperature) variation but suffers from a stiff challenge in achieving fast
lock. Phase-picking is an open-loop approach and enjoys the advantages of
a high data rate but are more sensitive to the effect of PVT variation.
In this laboratory, you are required to design a clock and data recovery
block using phase-tracking.
42
CHAPTER 6. CLOCK AND DATA RECOVERY 43
2 Laboratory Work
Reuse the serial link with the longest channel that you designed in the pre-
vious design laboratory. It is shown in Fig.6.1 for your convenience. The
recovered clock is the output of the oscillator of the PLL while the recovered
data is the output of the slicer.
Figure 6.1: Serial link. Additional circuitry (in red color) is added to compute
the BER.
3 Laboratory Report
Laboratory reports must be prepared in either Microsoft Word, LaTeX, or
other text editors. Figures, tables, and expressions should be embedded
in the main body of reports. All expressions, figures, and tables must be
numbered. All figures and tables must be captioned. Cadence print-outs
attached to the end of laboratory reports are not acceptable. Hand-
written laboratory reports will be rejected. Laboratory reports must
contain the followings :
• Table of contents
• List of figures
• List of tables
3. Conclusions (10%)
Provide no less than three (3) but no more than five (5) conclusions
that you can draw from this design laboratory. Your conclusions must
be supported by the simulation results obtained and must be as specific
and precise as possible.
Chapter 7
1 Introduction
The performance of DFE critically depends upon tap coefficients. The opti-
mal tap coefficients that yield the maximum eye openings varies with both
data rate and the characteristics of channels. In Laboratory 5, a trial-and-
error approach was used to find optimal tap coefficients. In this laboratory,
a systematic method that is based on the principle of least-mean-square will
be used to guide the search for optimal tap coefficients. The method searches
for optimal tap coefficients with the objective to minimize the power of the
difference between the desired and equalized data symbols, i.e., the output
and input of the slicer. In this design laboratory, you are required to design a
2-tap adaptive decision feedback equalizer that will utilize blocks such as the
transmitter, channel, continuous-time linear equalizer, and PLLs that you
designed in the previous design laboratories. The adaptive decision feedback
equalizer will consist of (i) a slicer that samples the difference between the
output of the continuous-time linear equalizer and the DFE taps, (ii) 2 unit
interval (UI) delay blocks whose propagation delay is the symbol time, (iii)
2 tap multipliers that multiply the past decisions of the slicer with appro-
46
CHAPTER 7. ADAPTIVE DECISION FEEDBACK EQUALIZERS 47
priate tap coefficients, (iv) a summer that subtracts the DFE taps from the
output of the continuous-time linear equalizer, and (v) an adaptive engine
that computes the amount of the adjustment of tap coefficients and adjusts
the tap coefficients by varying the tail currents of the tap multipliers.
The uncertainty of the characteristics of channels requires that tap co-
efficient be set in accordance of both data rate and the characteristics of
channels such that a complete removal of post cursors can be achieved. The
optimal tap coefficients are obtained by minimizing the power of the differ-
ence between the desired and equalized data symbols measured at the center
of the data eye, i.e., least-mean-square (LMS). In a LMS adaptive DFE, the
jth optimal tap coefficient is obtained using iterative operation
where cj,k+1 and cj,k are the coefficients of tap-j in iterations k + 1 and k,
respectively, k is the difference between the output and input of the slicer
measured at the center of the data eye, i.e., the desired and equalized data
symbols, and h is the step size used to adjust the tap coefficient.
The preceding LMS adaptive DFE is difficult to implement due to the
need for k and vk , which can only be obtained using ADCs. Sign-sign LMS
adaptive DFE that only uses the signs of k and vs,k rather than their actual
values to obtain the optimal tap coefficient
2 Laboratory Work
2.1 Serial Link with Adaptive DFE
Construct the serial link consisting of the PRBS generator, transmitter,
block, channel, CTLE, PLLs, and adaptive DFE designed previously as per
Fig.7.1. The core of the adaptive DFE is the adaptive engine that performs
the following tasks : (i) Compute the amount of the adjustment of tap co-
efficients. (ii) Adjust tap coefficients. The amount of the adjustment of tap
coefficients is given by hsign[k ]sign[vs,k ]. sign[vs,k ] is the output of the slicer
and is readily available. k quantifies the difference between the desired and
equalized data symbols. Although the desired data symbols are the output of
the slicer, if the output of the slicer is 1, the difference will always be greater
than 0. Similarly, if the output of the slicer is 0, the difference will always be
less than 0. These observations reveal that k should be defined differently
in order to ensure the proper operation. Since if the transmitted data is 0,
the equalized data should also be 0 ideally. Similarly, if the transmitted data
is 1, the equalized data should also be 1 ideally. These observations suggest
that k should be generated by comparing the equalized data with Vref,H if
the transmitted data is 1 and Vref,L if the transmitted data is 0 where Vref,H
and Vref,L are desired voltage for 1 and that for 0, respectively. The exact
values of Vref,H and Vref,L depend on both data rate and the characteristics
of the channel over which data are transmitted. A simple logic circuit can
therefore be developed to generate sign[k ]. The inputs to this logic circuit
include (i) the equalized data symbol, (ii) Vref,H , (iii) Vref,L , and (iv) the
output of the slicer that indicates whether the transmitted data is 0 or 1.
Although the amount of the adjustment of tap coefficients is given by
hsign[k ]sign[vs,k ]. sign[vs,k ], only h is an analog quantity whereas the other
two are digital, revealing that h quantifies the amount of the adjustment of
tap coefficients while sign[k ]sign[vs,k ] determines whether the adjustment
is incremental or decremental. A natural circuit that can provide both
incremental or decremental changes of a voltage is a charge pump. The
CHAPTER 7. ADAPTIVE DECISION FEEDBACK EQUALIZERS 49
Figure 7.1: Serial link with pre-emphasis and adaptive decision feedback
equalization
2.2 Adaptation
Perform transition analysis of the serial link and plot the output voltages
of the two charge pumps that set the tail currents of the tap multipliers.
When adaptation ends, the voltages of the charge pumps will converge to
constant values. These two plots show the adaptation process, in particular,
the amount of the time needed to find the optimal tap coefficients.
3 Laboratory Report
Laboratory reports must be prepared in either Microsoft Word, LaTeX, or
other text editors. Figures, tables, and expressions should be embedded
in the main body of reports. All expressions, figures, and tables must be
numbered. All figures and tables must be captioned. Cadence print-outs
attached to the end of laboratory reports are not acceptable. Hand-
written laboratory reports will be rejected. Laboratory reports must
contain the followings :
• Schematics (20%)
– The schematic of the adaptive engine.
– The complete schematic (block-level diagram) of the serial
link.
• The eye diagram of the serial link when adaptation has ended
(20%).
CHAPTER 7. ADAPTIVE DECISION FEEDBACK EQUALIZERS 51
• The plots of the output voltages of the two charge pumps showing
adaptation process (20%).
• The plots of the current drawn by (i) the transmitter, (ii) trans-
mitter PLL, (iii) CTLE, (iv) receiver PLL, and (v) adaptive DFE
and a table tabulating the average power consumption of these
blocks (20%).
3. Conclusions (10%)
Provide no less than three (3) but no more than five (5) conclusions
that you can draw from this design laboratory. Your conclusions must
be supported by the simulation results obtained and must be as specific
and precise as possible.