Undersampling in SDR (software defined radio)

Preliminary discussion

Software defined radio

A striking feature of the relatively short history of electronic communications is
the significant improvement in performance with each innovation (usually in terms of
bandwidth requirements and/or noise immunity). This has often meant that, as better
communications systems have been introduced, they have quickly replaced existing
technologies. For a recent example of this, consider the switch from analog to digital cell
phones.

However, where the existing technology has been too well established to be
abandoned, the new system has run in parallel with the old. For a long-standing example
of this, consider the commercial AM and FM radio systems.

Despite the benefits of new communications techniques, the disadvantages can’t

be ignored. Hardware is either rendered useless or it must be duplicated. These problems
have lead to the development of the latest communications concept called software
defined radio (SDR). SDR is a single tuner that can receive and decode any of the
existing communications formats (AM, FM, DSBSC, ASK, FSK, DSSS, etc). Moreover, it’s is
also capable of decoding any communications format that will be developed in the
foreseeable future.

As its name implies, the astounding flexibility of SDR is achieved using software.
Instead of implementing a hardware receiver that is necessarily band and modulation-
scheme specific, SDR is a wideband receiver that converts radio signals to digital then
decodes them using the software appropriate to the modulation scheme of the
transmission signal. For a different modulation scheme, simply change the program.
Better still, for a new modulation scheme, simply install the new program that’s capable
of decoding it.

Undersampling
An SDR receiver capable of receiving (and decoding) the majority of electronic
communications would need to operate at frequencies up to and beyond 2.4GHz (a
typical cell phone frequency). Recalling the Nyquist Sample Rate, you might be tempted
to imagine the SDR receiver’s Analog-to-Digital Converter (ADC) needing to sample cell
phone signals at over 4.8GHz! However, the Nyquist requirement to sample at two or
more times the highest frequency of the input signal is for avoiding aliasing of baseband
signals.

Bandwidth limited signals (like radio signals in communications) don’t have

frequency components near DC. That being the case, the type of aliasing that the
Nyquist Sample Rate attempts to avoid isn’t a problem. In fact, Shannon’s Information
Theorem states that all of the information in a bandwidth limited signal can be captured
with a sampling rate as low as twice the signal’s bandwidth.

In other words, a 2.4GHz carrier signal with a 30kHz bandwidth can be sampled at
a frequency as low as 60kHz and still capture all of the signal’s information. That said,
there are certain sampling frequencies that will still cause aliasing and there is a
mathematical process for identifying them.

Sampling of bandwidth limited signals at less than the Nyquist Sample Rate is
known as undersampling, band-pass sampling and super-Nyquist sampling. Importantly,
as well as allowing for communications signals up to very high frequencies to be
sampled, undersampling has another significant advantage that makes it ideal for SDR.
When the undersampling frequency is twice the signal’s bandwidth, one of the sampled
signal’s aliases occurs at the same frequency as the original message used to modulate
it. In other words, undersampling demodulates the sampled signal. All that need be done
to recover the original message is to pass it through a low-pass filter to filter out the
higher frequency aliases.

The experiment
In this experiment you’ll use the Emona Telecoms-Trainer 101 to set up a
bandwidth limited signal. You’ll then use undersampling to demodulate the bandwidth
limited signal and recover the message. Finally, you’ll explore the effects on the
recovered message of mismatches between the modulated carrier’s bandwidth and the
frequency used for undersampling.

It should take you about 40 minutes to complete this experiment.

Part A – Setting up a bandwidth limited signal

To experiment with undersampling you need a bandwidth limited signal. Any of
the modulation schemes can be used for this purpose, but for simplicity of wiring, we’ll
use a DSBSC signal. The first part of the experiment gets you to set one up.

Procedure

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

2. Set up the scope per the instructions in Experiment 1. Ensure that:

 the Trigger Source control is set to the CH1 (or INT) position.
 the Mode control is set to the CH1 position.

3. Connect the set-up shown in Figure 1 below.

M A S T ER M U LTIP L IER
S IGN A L S

DC
X
AC

DC
Y
1 0 0 kH z
AC
S IN E
1 0 0 kH z kX Y
C OS
M U LTIP L IER
1 0 0 kH z
D IGITA L
8 kH z
D IGITA L
2 kH z X DC
D IGITA L
2 kH z
S IN E
Y DC kX Y

Figure 1

This set-up can be represented by the block diagram in Figure 2 below. It

generates a 100kHz carrier that is DSBSC modulated by a 2kHz sinewave message.

M essage
T o Ch.1
M ast er M ult iplier
S ignals module
Y
D S BS C signal
2 kH z T o Ch.2
X
10 0 kH z
car r ier

M ast er
S ignals

Experiment 20 – Undersampling in software defined radio

20-2 © Emona Instruments
 Figure 2

4. Adjust the scope’s Timebase control to view two or so cycles of the Master Signals
module’s 2kHz SINE output.

5. Set the scope’s Mode control to the DUAL position to view the DSBSC signal out of
the Multiplier module as well as the message signal.

6. Set the scope’s Channel 1 Vertical Attenuation control to the 1V/div position and
the Channel 2 Vertical Attenuation control to the 2V/div position.

Note: The Multiplier module’s output should be DSBSC signal with alternating
halves of its envelope forming the same shape as the message.

Question 1
For the given inputs to the Multiplier module, what are the frequencies of the two
sinewaves that make up the DSBSC signal?

Question 2
What’s the bandwidth of the DSBSC signal?

Part B – Direct down-conversion using undersampling

If you have successfully completed the experiment on sampling and reconstruction
(Experiment 11) you have seen that the mathematical model that defines the sampled
signal is:

Sampled signal = the sampling signal × the message

As the sampling signal is a digital signal, the expression can be rewritten as:

Sampled signal = (DC + fundamental + harmonics) × message

When the message signal is modulated carrier like the DSBSC signal that you
have set up, the expression can be rewritten as:

Sampled signal = (DC + fundamental + harmonics) × (LSB + USB)

Solving the expression (which necessarily involves trigonometry that is not shown
here) gives:

 Duplicates of the LSB and USB (due to their multiplication with sampling signal’s DC
component)
 Aliases of the LSB and USB at frequencies equal to the sum and difference of their
frequencies and the sampling signal’s fundamental frequency
 Numerous other aliases of the LSB and USB at frequencies equal to the sum and
difference of their frequencies and the sampling signal’s harmonic frequencies

Recall that the math also proves that, where a low-pass filter is being used to
reproduce the original signal by plucking its equivalent out of the sampled signal, the

Experiment 20 – Undersampling in software defined radio

20-3 © Emona Instruments
sampling rate must be at least twice the highest frequency in the original signal. If the
sampling rate is less than this, aliasing occurs.

At first glance then, this suggests that if the DSBSC signal that you have generated is
to be sampled, the sampling rate must be at least 204kHz because of the upper sideband
is a 204kHz sinewave.

However, as the DSBSC signal is bandwidth limited (that is, its spectral composition
doesn’t extend down to DC), it’s possible to sample at rates lower than 204kHz without
necessarily causing aliasing. For proof, Table 1 on the next page shows some of the
aliases produced by sampling the DSBSC signal at 150kHz.

Table 1
Components due Components due Components due Components due
to DC to fs to 2fs to 3fs

Notice that none of the aliases overlap the 98kHz and 102kHz components in the
sampled signal’s spectral composition. The aliases are either below or above them. So, in
this instance, aliasing wouldn’t occur if a band-pass filter (with sufficiently steep skirts) is
used to pluck the duplicate of the original DSBSC signal out of the sampled signal. That
said, aliasing is still possible by choosing a sampling rate that produces aliases at
frequencies that fall inside the band-pass filter’s pass-band.

Obviously, as the sampling rate decreases, so too do all of the components in the
sampled signal’s spectrum. It makes sense then that, if the right undersampling
frequency is used, it must be possible to produce aliases centre on DC. This is crucial
because it means that, when a modulated carrier is undersampled, one of its sidebands
can be directly down-converted back to a baseband signal without needing to use an
intermediate frequency first. All that is needed is a low-pass filter to reject the other
aliases.

A more sophisticated way of understanding direct down-conversion using

undersampling involves thinking of the sampling action as product detection. This is
entirely appropriate to do because the math is almost identical – if you’re not sure about
that, compare the notes here with the notes in the preliminary discussion on product
detection in Experiment 7. The difference is however, instead of multiplying the
modulated carrier with a single local sinusoidal carrier, sampling involves multiplying it
with dozens of sinewaves (the sampling signal’s fundamental and harmonics).
Importantly, as long as one of the harmonics is the same frequency as the modulated
carrier, the explanation for a product detector applies equally to undersampling as a
form of demodulation.

To ensure that one of the sampling signal’s harmonics is the same frequency as
the modulated carrier, the sampling rate must be a whole integer sub-multiple of the
modulated signal’s carrier frequency. That said, to avoid aliasing, the sampling rate must
be at least twice the bandwidth limited signal’s bandwidth.

The next part of this experiment lets you demodulate your DSBSC signal to
recover the 2kHz message using undersampling instead of using a product detector.

7. Return the scope’s Channel B Scale control to the 0.5V/div position.

8. Modify the set-up as shown in Figure 3 on the next page.

Experiment 20 – Undersampling in software defined radio

20-4 © Emona Instruments
 M A S TER M U LTIP L IER D U A L A N A L OG C H A N N EL
S IGN A L S S W IT C H M OD U L E
S/ H

DC
X S& H S& H C H A N N EL
AC IN OU T BPF

DC
Y IN 1 B A S EB A N D
1 0 0 kH z AC
S IN E LP F

1 0 0 kH z kX Y
C OS
M U LTIP L IER A D D ER
1 0 0 kH z C ON TR OL 1
D IGITA L C ON TR OL 2
8 kH z N OIS E
D IGITA L
2 kH z X DC
D IGITA L
2 kH z S IGN A L C H A N N EL
S IN E OU T
Y DC kX Y IN 2 OU T

Figure 3

This set-up can be represented by the block diagram in Figure 4 below. The Multiplier
module is used to generate a modulated carrier (DSBSC). The Sample-and-Hold circuit
together with the Baseband LPF is used demodulate it using undersampling.

U nder - sampled
M essage D S BS C signal
T o Ch.1 T o Ch.2

Baseband
L PF

Y IN Recover ed
S/ H message
2 kH z
X CO N T RO L
10 0 kH z
car r ier 8 kH z

M ast er
S ignals

D S BS C modulat or D emodulat ion

9. Compare the undersampled DSBSC signal with the original message.

Note: If you look closely, the undersampled DSBSC signal looks a little like an
inverted version of the original message.

Figure 4

10 Modify the scope’s Channel B connection to the set-up as shown in Figure 5

. below.

Experiment 20 – Undersampling in software defined radio

20-5 © Emona Instruments
 M A S TER M U LTIP L IER D U A L A N A L OG C H A N N EL
S IGN A L S S W IT C H M OD U L E
S/ H

DC
X S& H S& H C H A N N EL
AC IN OU T BPF

DC
Y IN 1 B A S EB A N D
1 0 0 kH z AC
S IN E LP F

1 0 0 kH z kX Y
C OS
M U LTIP L IER A D D ER
1 0 0 kH z C ON TR OL 1
D IGITA L C ON TR OL 2
8 kH z N OIS E
D IGITA L
2 kH z X DC
D IGITA L
2 kH z S IGN A L C H A N N EL
S IN E OU T
Y DC kX Y IN 2 OU T

Figure 4

Question 3
What’s the significance of the signal on the Baseband LPF’s output?

Question 4
Given the sampling frequency is 8.333kHz (the signal’s specified value of 8kHz is
rounded down for simplicity), which harmonic in the sampling signal is
demodulating the DSBSC signal?

Part C – Synchronisation
Recall that transmitter and receiver carrier synchronisation is essential to
successful demodulation using product detection. If the local carrier of a product detector
has even the slightest frequency or phase error (relative to the modulated carrier), the
demodulated signal is affected.

Phase errors can reduce the magnitude of the recovered message and even result
its complete cancellation. The effect of frequency errors depends on size. If the error is
small (say 0.1Hz) the message is periodically inaudible but otherwise intelligible. If the
frequency error is larger (say 5Hz) the message is reasonably intelligible but fidelity is
poor. When frequency errors are large, intelligibility is seriously affected. (For a brief
explanation of why these effects occur, refer to Part E in Experiment 7.)

As direct down-conversion using undersampling is a form of product detection,

the sampling signal must be synchronised to the modulated carrier if these effects are to
be avoided. The next part of the experiment let’s you see these effects for yourself.

Experiment 20 – Undersampling in software defined radio

20-6 © Emona Instruments
 11. Locate the VCO module and set its Range control to the LO position.

12. Set the VCO module’s Frequency Adjust control to about the middle of its travel.

13. Set scope’s Mode control so that only Channel 1 is displayed.

14. Connect the scope’s Channel 1 input to the VCO module’s Digital output.

15 Adjust the VCO module’s Digital output to 8.333kHz.

.
1
P=
Note: You do this by adjusting the signal’s period to 120µs (recall that f ).

16. Return the scope’s Channel 1 input to the Master Signals module’s 2kHz SINE
output.

17 Readjust the scope’s controls to view two or three cycles of both the original and
. recovered messages.

18 Disconnect the plug to the Master Signal module’s 8kHz DIGITAL output.
.

19 Modify the set-up as shown in Figure 6 below.

.
Note: Leave the scope’s connections as they are.

VCO M A S T ER M U LT IP L IER D U A L A N A L OG C H A N N EL
S IGN A L S S W IT C H M OD U L E
S/ H

D IGITA L
DC
X S& H S& H C H A N N EL
AC IN OU T BPF

GA IN DC
Y IN 1 B A S EB A N D
1 0 0 kH z AC
S IN E LP F

1 0 0 kH z kX Y
F R EQ C OS
M U LT IP L IER A D D ER
1 0 0 kH z C ON TR OL 1
D IGITA L C ON TR OL 2
HI N OIS E
8 kH z
LO D IGITA L
2 kH z X DC
D IGITA L
VC O S IN E 2 kH z S IGN A L C H A N N EL
IN P U T S IN E OU T
Y DC kX Y IN 2 OU T

Figure 5

This modification substitutes the Master Signals module’s 8kHz DIGITAL output
(which is actually 8.333kHz) for an 8.333kHz digital signal from the VCO module.
However, it is highly unlikely that the VCO module’s output will be exactly the right
frequency at exactly the right phase for successful demodulation.

20. Observe the effect of this change on the recovered message.

21. Try to correct the frequency error by turning the VCO module’s Frequency Adjust
control very slightly to the left and right.

Question 5
Why doesn’t this solve the problem and allow the demodulator to recover the
message?

Experiment 20 – Undersampling in software defined radio

20-7 © Emona Instruments
Experiment 20 – Undersampling in software defined radio
20-8 © Emona Instruments