Pulse code modulation and signal-to-noise distortion ratio (SNDR)

Preliminary discussion

PCM encoding and decoding on the Emona Telecoms-Trainer 101 is like all other
digitisation schemes in that it suffers from quantisation error. To explain, analog-to-digital
conversion involves sampling the input voltage (which will be one of an infinite number of values
between the range) and comparing it to a set of quantisation levels. As the number of
quantisation levels is finite, the conversion process must allocate the quantisation level that is
the closest to the sample and generate the number that represents it. Immediately upon doing
so, the actual size of the sample is lost and this is quantisation error. As it’s impossible for the
PCM decoder to determine the difference between the actual size of the sample and value that
the PCM encoder inferred it to be, the PCM decoder reproduces the error on its output.

Importantly, there is a greater chance of a sampled analog voltage being different to

one of the quantisation levels than the same. In other words, this error is almost always present
and distorts the reproduction of the original signal on the PCM decoder’s output. Moreover, the
spectral composition of this distortion is similar to the spectral composition of white noise and
so quantisation error is said to cause quantisation noise. In fact, when audible, quantisation noise
can be heard as hiss (provided the quantisation error isn’t too large).

Recall that improving a digitisation scheme’s resolution involves increasing the number of
bits in the binary number. This in turn increases the number of quantisation levels permissible
and so the quantisation error involved in every sample must be smaller. Clearly then, quantisation
noise is a function of the scheme’s resolution. This gives us a potential opportunity to calculate
the theoretical size of the quantisation noise and express it as a signal-to-noise distortion ratio
(SNDR).

However, there are three problems with doing so. First, as the quantisation error is
fixed (at ±½ bit) any calculated SNDR figure is a function of the size of the analog signal. The
bigger the analog signal, the bigger the largest binary number it produces and so the better the
SNDR (that is, ±½ bit is a much smaller proportion of 11111111 than it is of 00000001). While
this isn’t problematic in itself, it is when the message is speech which doesn’t have fixed
amplitude. Second, SNDR is affected by other factors like timing errors between the clocks of
the analog-to-digital and digital-to-analog converters. These can be difficult to predict and
incorporate into any theoretical calculations of SNDR. Third, PCM systems invariably use a
reconstruction filter to remove the aliases (or images) from the decoded PCM data signal. This
filtering inevitably removes some of the noise also with the amount of noise rejection being a
function of the filter’s performance.

These factors combine to make measurement of SNDR an important part of determining

the quality of a digitisation scheme for a specific design.
 Determining SNDR by measurement is similar to determining conventional signal-to-noise
ratio (SNR) by measurement. This typically involves measuring the RMS noise voltage (N)
independently of the signal then measuring either the RMS value of the recovered message
signal voltage (S) or the RMS value of the recovered message signal plus noise voltage (S+N).
RMS values are used so that the SNR figure is an average for the changing size of the message.
The SNR is then calculated using either:

S S+ N
SNR = SNR =
N or N

However, there are couple of small catches when doing this for SNDR. First, it’s not
possible to measure the recovered message signal without quantisation noise because
quantisation noise is always present – it’s a property of digitisation. This means that the second
of the two equations above must be used.

Second, it not possible to measure the quantisation noise independent of the signal in
the same way as for conventional SNR. When you think about this, it makes sense. Without an
analog input signal there are no quantisation errors to cause the quantisation noise in the first
instance. This means that additional processing is needed prior to measurement to isolate the
noise from the signal.

A method for doing this involves a technique of nulling signals that was introduced to you
in Experiment 4 of Volume 1. Figure 1 below shows the block diagram of how this is implemented.

M essage plus
O utput
quantisation noise

"S tolen" m essage

O
Figure 1

The signal with its quantisation noise is fed to one input of an adder. A copy of the
original message (stolen from the source) is connected to the adder’s other input via a phase
shifter. The phase shifter is then adjusted so that the stolen message is 180° out of phase with
the recovered message. Once done, the adding process cancels the two messages completely
(ideally) leaving only quantisation noise to be measured.

Experiment
10-2 10 – Pulse code modulation and SNDR © Emona Instruments
The experiment
For this experiment you’ll use the Emona Telecoms-Trainer 101 to set up a PCM communications
system. You’ll then use it to determine the SNDR of the signal on the PCM decoder’s output
(that is, the signal prior to reconstruction filtering). Next, you’ll complete the PCM system by
adding a reconstruction filter using a simple RC low-pass filter and measure its SNDR. Finally,
you’ll substitute the RC LPF for a more sophisticated filter and examine the effect this has on
SNDR.

It should take you about 50 minutes to complete this experiment.

Procedure

Part A – Setting up the PCM encoding-decoding scheme

To measure the signal-to-noise distortion ratio (SNDR) of a PCM encoding-decoding scheme, you
must first set one up. Part A gets you to do this.

1. Gather a set of the equipment listed on the previous page.

2. Set the PCM Encoder module’s Mode switch to the PCM position.

3. Connect the set-up shown in Figure 2 below.

Figure 2

M AS TER P CM P CM
SIGNALS ENCODER DECODER

P CM TDM

TDM

1 0 0 kHz
S IN E
1 0 0 kHz INP UT 2 FS FS
COS
1 0 0 kHz
DIGITAL
8 kHz INP UT 1 P CM OUTP UT2
DIGITAL DATA

2 kHz
DIGITAL
2 kHz CLK P CM CLK OUTP UT
S IN E DATA

The set-up in Figure 2 can be represented by the block diagram in Figure 3 below
M essage
T o CH 1
M aster
S ignals FS

D ecoded
PCM data
2 kH z IN
PC M To CH 2
C LK DATA C LK

10 0 kH z

M aster
S ignals

PC M E ncod ing PC M D ecoding

Experiment
10-3 10 – Pulse code modulation and SNDR © Emona Instruments
 Figure 3

The Master Signals module’s 2kHz SINE output is the message. The PCM Encoder
module converts the message to a digital signal which inevitably contains quantisation errors.
The PCM Decoder module returns the digital signal to a sampled version of the original signal
consisting of quantisation noise and aliases.

4. Set up the scope per the instructions in the Appendix.

Note: Ensure that the scope’s Trigger Source control is set to the CH 1 (or INT)
position.

5. Adjust the scope’s Timebase control to view two or so cycles of the message.

6. Set the scope’s Mode control to the DUAL position to observe the signal on the PCM
Decoder module’s output as well as the message.

Note: The PCM Decoder module’s output should be a sampled version of the message.

Part B – Measuring the SNDR of the signal on the PCM decoder’s output
To determine the signal-to-noise distortion ratio (SNDR) of the signal on the PCM
decoder’s output by measurement, you must first measure its RMS voltage.

7. Set up the True RMS digital multimeter (DMM) for measuring AC voltages.

Note: If you’re not sure how this is done, call the instructor for assistance.

8. Connect the True RMS DMM between the PCM Decoder module’s output and the
trainer’s ground.

9. Measure the RMS voltage of the PCM Decoder module’s output and record this value in
Table 1 on page 10-11.

Next, you must measure the RMS value of the noise on its own. This involves nulling the
message component in the signal on the PCM decoder’s output and the next part of the
experiment gets you to do this.

10. Locate the Adder module and set its G and g controls to about the middle of their
travel.

11. Locate the Phase Shifter module and set its Phase Change control to the 0° position.

12. Set the Phase Shifter module’s Phase Adjust control about the middle of its travel.

13. Modify the set-up as shown in Figure 4 on the next page.

Experiment
10-4 10 – Pulse code modulation and SNDR © Emona Instruments
 M AS TER P CM P CM
SIGNALS EN CODER DECODER

P CM TDM

TDM

1 0 0 kHz
S IN E
1 0 0 kHz IN P UT 2 FS FS
COS
1 0 0 kHz
DIGITAL
8 kHz IN P UT 1 P CM OU TP UT2
DIGITAL DATA

2 kHz
DIGITAL
2 kHz CLK P CM CLK OU TP UT
S IN E DATA

P HASE ADDER
S HIFTER

LO

P HAS E
G
O
0
A
O
180

IN OUT g

B GA+g B

Figure 3

The additions to this set-up can be represented by the block diagram in Figure 5 on the
next page. The 2kHz message is stolen from the Master Signals module and phase shifted using
the Phase Shifter module. The signal is then added to the signal on the PCM Decoder module’s
output using the Adder module.

A dde r

A
D e coded PCM signal N oise
To CH 2
B

O Phase
S hifte r

"S tole n" 2 kH z

m essage

Figure 4
The entire set-up can be represented by the block diagram in Figure 6 below.

Experiment
10-5 10 – Pulse code modulation and SNDR © Emona Instruments
 M essage
To CH 1

FS

A
N oise
2kH z IN T o CH 2
PC M
CLK DATA CLK B

100 kH z

O
PC M E ncoding PCM D ecoding

"S tolen" 2kH z

m essage
Figure 5

14. Disconnect the lead to the Adder module’s B input.

15. Adjust the Adder module’s G control until its output is as close to 4Vp-p as you can
achieve.

16. Reconnect the lead to the Adder module’s B input.

17. Disconnect the lead to the Adder module’s A input.

18. Adjust the Adder module’s g control until its output is as close to 4Vp-p as you can
achieve.

Note: Once done, the gain of the Adder module’s two inputs are close to equal.

19. Reconnect the lead to the Adder module’s A input.

20. Vary the Phase Shifter module’s Phase Adjust control left and right.

Note: As you do this, you’ll vary the phase relationship between the signal on the PCM
decoder’s output and the stolen message. Importantly, when the phase relationship
between the stolen message and the message component of the decoded PCM signal is
180°, they cancel each other leaving only noise.

21. Connect the True RMS DMM between the Adder output and the trainer’s ground.

22. Adjust the Phase Shifter module’s Phase Adjust control until you obtain the smallest
RMS output voltage from the Adder module.

Experiment
10-6 10 – Pulse code modulation and SNDR © Emona Instruments
 To obtain the most accurate measurement of noise, the recovered message must be
nulled as much as the set-up will allow. This requires a little fine-tuning of the current Adder
and Phase Shifter module settings. The next few steps get you to do this.

23. Vary the Adder module’s G control just a little to see if you can make its output smaller.

24. Adjust the Phase Shifter module’s Phase Adjust control just a little to see if you can
make the Adder module’s output smaller.

25. Repeat Steps 23 and 24 until you obtain the smallest possible output from the Adder
module.

26. Once you have nulled the message component on the PCM decoder’s output as much as
possible, record the RMS “noise” voltage in Table 1 on the next page.

The next part of the experiment gets you to calculate the SNDR of an 8-bit PCM encoding-
decoding scheme.

27. Calculate and record the SNDR figure using the equation:

S+N
SNDR=
N

28. Convert the SNDR figure to decibels and record this also.

Table 1
Signal on the PCM
decoder’s output

Noise

SNDR

SNDR
(in decibels)

Question 1
What factor limits the accuracy of this measurement?

Question 2
What two factors contribute to the noise component that you measured?

Experiment
10-7 10 – Pulse code modulation and SNDR © Emona Instruments
Part C – Measuring the SNDR of a complete PCM system

PCM encoding-decoding systems are unlikely to use the signal on the PCM decoder’s
output without further conditioning because, apart from the quantisation noise, it contains
components not in the original message (the aliases or images). These may be audible and so
filtering is usually used to reconstruct the message by rejecting the aliases. Usefully, the
filtering process also filters out some of the quantisation noise. Part C of the experiment gets
you to complete the PCM system by adding filters of differing grades and measuring their
effect on SNDR.

29. Modify the set-up as shown in Figure 7 below.

Note: Disconnect the patch lead to the Adder module’s A input and leave it floating for
a moment.

M AS TER P CM P CM UTILITIES
S IGNALS EN CODER DECODER
COM PARATOR
REF

P CM TDM

TDM

1 0 0 kH z IN OUT
S IN E RECTIFIER
1 0 0 kH z IN P UT 2 FS FS
COS
1 0 0 kH z
DIGITAL DIODE & RC LP F
8 kHz IN P UT 1 P CM OUTP UT2
DIGITAL DATA

2 kHz
DIGITAL RC LP F
2 kHz CLK P CM CLK OUTP UT
S IN E DATA

P HAS E ADDER
S HIFTER

LO

P HAS E
G
O
0
A

O
180

IN OUT g

B GA+g B

Figure 6

Experiment
10-8 10 – Pulse code modulation and SNDR © Emona Instruments
 The set-up in Figure 7 can be represented by the block diagram in Figure 8 below. The
RC LPF module is used to reconstruct the message by filtering.

M essage
To CH 1

FS R C LPF

R ecovered
m essage
2 kH z IN
PC M To CH 2
C LK DATA C LK

10 0 kH z

PCM En coding PC M D ecoding R econstruction

Figure 7

30. Compare the original message and the reconstructed message.

Note: The signals should be similar but the reconstructed message will show obvious
signs of distortion as not all of the aliases are removed.

Experiment
10-9 10 – Pulse code modulation and SNDR © Emona Instruments
 31. Connect the True RMS DMM between the RC LPF’s output and the trainer’s ground.

32. Measure the RMS voltage on the RC LPF’s output and record this value in Table 2 on
page 10-16.

33. Modify the set-up as shown in Figure 9 below.

M AS TER P CM P CM UTILITIES
S IGN ALS EN CODER DECODER
COM PARATOR
REF

P CM TDM

TDM

1 0 0 kHz IN OUT
S IN E RECTIFIER
1 0 0 kHz IN P UT 2 FS FS
COS
1 0 0 kHz
DIGITAL DIODE & RC LP F
8 kHz IN P UT 1 P CM OUTP UT2
DIGITAL DATA

2 kHz
DIGITAL RC LP F
2 kHz CLK P CM CLK OUTP UT
S IN E DATA

P HAS E ADDER
S HIFTER

LO

P HAS E
G
O
0
A

O
180

IN OUT g

B GA+g B

Figure 8

This set-up can be represented by the block diagram in Figure 10 on the next page. The nulling
circuit has been re-introduced to remove the component on the filter’s output that is at the
message frequency leaving only the noise (ideally).

Experiment
10-10 10 – Pulse code modulation and SNDR © Emona Instruments
Experiment
10-11 10 – Pulse code modulation and SNDR © Emona Instruments
 M essage
To CH 1

FS

A
N oise
2 kH z IN To CH 2
PCM
CLK DA TA C LK B

10 0 kH z

O
PC M E ncod ing PC M D ecoding Reconstruction

"S tolen" 2 kH z
m essage
Figure 9

34. Connect the True RMS DMM between the Adder output and the trainer’s ground.

35. Adjust the Phase Shifter module’s Phase Adjust control until you obtain the smallest
output from the Adder module.

36. Tweak the Adder module’s G control to see if you can make its output smaller.

37. Tweak the Phase Shifter module’s Phase Adjust control to see if you can make the
Adder module’s output even smaller.

38. Once you have nulled the message component on the reconstruction filter’s output as
much as possible, record the RMS “noise” voltage in Table 2 below.

39. Calculate and record the SNDR figure as a ratio and in decibels.

Table 2
Signal plus
noise

Noise

SNDR

SNDR
(in decibels)

Question 3
How does this SNDR figure compare to that for the PCM system without reconstruction
filtering?

Question 4

Experiment
10-12 10 – Pulse code modulation and SNDR © Emona Instruments
 Explain why this change in SNDR has occurred.

Importantly, an RC LPF is a relatively simple filter to implement and as such doesn’t have
the performance characteristics of more sophisticated filter designs. This type of filter would
be used for message reconstruction of PCM systems in products for sale in price-sensitive
markets such as intercoms. However, the domestic audio and pro-audio markets are not quite as
price-sensitive and so more sophisticated and expensive filtering can be used. The next part of
the experiment gets you to measure the effect on SNDR of better filtering.

40. Locate the Tuneable Low-pass Filter module and adjust its Gain control for maximum
gain.

41. Adjust the Tuneable Low-pass Filter module’s Cut-off Frequency Adjust control for the
highest cut-off frequency.

42. Modify the set-up as shown in Figure 11 on the next page.

Experiment
10-13 10 – Pulse code modulation and SNDR © Emona Instruments
 M AS TER P CM P CM TUN EABLE
S IGN ALS EN CODER DECODER LP F

P CM f Cx1 0 0
TDM

TDM

1 0 0 kHz
S IN E
1 0 0 kHz IN P UT 2 FS FS
COS fC
1 0 0 kHz
DIGITAL
8 kHz IN P UT 1 P CM OUTP UT2
DIGITAL DATA

2 kHz
DIGITAL GAIN
2 kHz CLK P CM CLK OUTP UT
S IN E DATA
IN OUT

P HAS E ADDER
S HIFTER

LO

P HAS E
G
O
0
A
O
180

IN OUT g

B GA+g B

Figure 10

This set-up can be represented by the block diagram in Figure 12 on the next page. The
Tuneable Low-pass Filter module is now being used to reconstruct the message.

Experiment
10-14 10 – Pulse code modulation and SNDR © Emona Instruments
 M essage
T o CH 1
T uneable
FS Low -pass Filter

R ecovered
m essage
2kH z IN
PCM T o CH 2
CLK DATA CLK

10 0kH z

PCM Encoding PCM D ecoding R econstruction

Figure 11

43. Compare the original message and the reconstructed message.

44. Slowly reduce the Tuneable Low-pass Filter module’s cut-off frequency until the
message has been recovered (ignoring phase shift).

45. Continue to slowly reduce the Tuneable Low-pass Filter module’s cut-off frequency
stopping when its output voltage just begins to attenuate.

Note: This sets the filter’s cut-off frequency to very near 2kHz effectively placing the
reconstructed message’s aliases deep into the filter’s stop band.

46. Connect the True RMS DMM between the Tuneable Low-pass Filter module’s output and
the trainer’s ground.

47. Measure the RMS voltage on the Tuneable Low-pass Filter module’s output and record
this value in Table 3 on page 10-21.

Experiment
10-15 10 – Pulse code modulation and SNDR © Emona Instruments
 48. Modify the set-up as shown in Figure 13 below.

M AS TER P CM P CM TUN EABLE

S IGN ALS EN CODER DECODER LP F

P CM f Cx1 0 0
TDM

TDM

1 0 0 kHz
S IN E
1 0 0 kHz IN P UT 2 FS FS
COS fC
1 0 0 kHz
DIGITAL
8 kHz IN P UT 1 P CM OUTP UT2
DIGITAL DATA

2 kHz
DIGITAL GAIN
2 kHz CLK P CM CLK OUTP UT
S IN E DATA
IN OUT

P HAS E ADDER
S HIFTER

LO

P HAS E
G
O
0
A
O
180

IN OUT g

B GA+g B

Figure 12

This set-up can be represented by the block diagram in Figure 14 on the next page. The
nulling circuit has been re-introduced to remove the component on the filter’s output that is at
the message frequency leaving only the noise (ideally).

Experiment
10-16 10 – Pulse code modulation and SNDR © Emona Instruments
 M essage
To CH 1

FS

A
N oise
2 kH z IN To CH 2
PCM
CLK DA TA C LK B

10 0 kH z

O
PC M E ncod ing PC M D ecoding Reconstruction

"S tolen" 2 kH z
m essage
Figure 13

49. Connect the True RMS DMM between the Adder output and the trainer’s ground.

50. Adjust the Phase Shifter module’s Phase Adjust control then the Adder module’s G
control to obtain a small output from the Adder module.

Note: You may find it more convenient to set the Phase Shifter module’s Phase Change
control to the 180° position.

51. Tweak the Phase Shifter module’s Phase Adjust control and the Adder module’s G
control to see if you can make the Adder module’s output even smaller.

52. Once you have nulled the message component in the reconstruction filter’s output as
much as possible, record the RMS “noise” voltage in Table 3 below.

53. Calculate and record the SNDR figure as a ratio and in decibels.

Table 3
Signal plus
noise

Noise

SNDR

SNDR
(in decibels)

Question 5
How does this SNDR figure compare to the SNDR when the reconstruction filter was
implemented using an RC LPF?

Question 6

Experiment
10-17 10 – Pulse code modulation and SNDR © Emona Instruments
 What specific characteristic of the Tuneable Low-pass Filter module’s frequency
response is much better than that of the simple RC LPF?

Experiment
10-18 10 – Pulse code modulation and SNDR © Emona Instruments