0% found this document useful (0 votes)
973 views5 pagesArduino Based SDR Panadapter Project
Arduino Based SDR Panadapter Project
Uploaded by
Benjamin DoverCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
973 views5 pagesArduino Based SDR Panadapter Project
Arduino Based SDR Panadapter Project
Uploaded by
Benjamin DoverCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 5
Ea
Technical Feature
A soundcard-less SDR panadapter
based on the Arduino
Live spectrum and waterfall display without a PC
INTRODUCTION. Over the years, | have
experimented with several types of software
defined radio receivers including the
SoftRock series and the Finningley SOR.
‘These provide | and Q output signais that
are usually processed and displayed using
2 PC and its associated soundcard, the
software to do this being freely available
(on the web [1]. Having recently purchased
‘a KX3, | sought an inexpensive route to
‘acquiring an accompanying standalone
ppanadapter and so firs looked at the design
by AAGBE, which is based on the Raspberry
ror Beagletione boards (21. This design
‘again follows the traditional route of using
‘a soundcard, withthe Pi or Beaglebone
dealing mainly with information display.
‘though Ihave been developing computer
software for 50 years (I wrate my fist code
in 1964, using Ferranti Mercury Autocode),
| ound the Pi route frustrating, to say the
least! Having worked my way through an
eleven page guide to setting up the ABE
softwere, | found that it didn’t run because
the P's OS has been changed and there
was now a missing dependency between
the various software libraies. A search
Of the P's Forum pages did not reveal 2
solution to this problem so I decided instead
to-see what | could do with an Arduino
based system that was entirely sotware-
based, leit did not require a soundcard,
Itis worth pointing out thatthe resulting
hardware will only cisplay signals; unlike
some PC and soundeard-based systems, it
will nat demodulate them, primarily because
(of processor and memory limitations,
Nevertheless, it can still provide useful
information to an operator in terms of band
‘occupancy and may also be capable of use
28a data logger for propagation monitoring
HARDWARE. The success
ofa project such as this
depends crucially on the
choice of computer board
and displey module.
Superficialy, the
Raspberry Pi might seem to
win hands down over the
Arduino because of its much
faster processor; however,
Pi software is normally
written in an interpretive
language, Python, whereas
the Arduino uses @ compiled
version of C++. When
feferrng to the Arduino,
‘most people normally think
ofthe ubiquitous Uno
board but its performance
is inadequate for this
project, not only in terms of
processor speed (16MHz2)
and working memory (2k8
SRAM) but also in terms of available 10
pins (14). Instead, | used an Arduino Due
board, which has an Atmel SAM3XBE ARM
Cortex-M3 32 bit processor running at
84MbHz, together with 96kb of SRAM and
a tolal of 54 10 pins. A Due board can also
do 12 bit ADC sampling, yet is available on
eBay for less than £20.
The 5 inch TFT colour display merule
used in the panadapter has 800x480 RGB
pixels, a touch screen and even a SD card
port, Nevertheless, it too is available via
eBay for less than £20. The module has a
standard 40 pin header that is also used
on several smaller and cheaper displays
advertised on eBay but these typically
hhave only 320x240 pine’. inital software
evelopment was done using one of the
PHOTO 1: Breadboard prototype of Arduino based panadapter.
we cay
wae LT Le
Dusted x,
cel fr ne
Power
FAURE 1: lek aga debs pret
smaller displays but because of the common
pinout ofthe header, it was 2 simple matter
to upgrade the display when the feasibility
ofthe project was confirmed.
A block ciagram of the panadapter is
shown in Figure 1. As well asthe Due
board and TFT display, the only other
hardware needed comprises single stage
high gain audio amplifiers for each ofthe
| and Q channels and their associated
software-switchable active low pass filters
The latter are needed to reduce aliasing
errors in the sampled | and Q data
Figure 2 shows the circuit ofthe | and
Q input ampiters, each of which provides
2 nominal gain of about 50. Figure 3
shows the generic circuit of one ofthe 8
pole active low pass fiters. In total, four
ofthese are needed: two for the I data
channel and two for the Q data channel
‘The wideband its have a nominal cut-off
frequency of about 20kH2, the narrowband
‘ones about 4kH2, The cutoff frequencies
are set by changing the values of several
Capacitors and resistors (3] and the
appropriate values for each filter are listed
in Table 1. The signal paths through the
aporopriat filers are switched by software
Controlled relays using the circuit shown in
Figure 4. If required, the individual | and
Q channel gains can be set by multi-turn
potentiometers connected between the
‘outputs ofthe active low pass filters and
23 2a
Technical Feature
PHOTO 2: The panadapter displaying signals from the KX3 receiver
tuned to 10.1407 5MH2 (bandwidth set at =5
the I and Q ADC inputs on the Due boar.
‘The connections between the latter and
the TFT display are shown in Table 2.
The display backlight is powered via the
LED A connection, pin 19, on the 40 pin
header. Preferably this should be connected
toa separate 3.3V supply, but in practice
the on-board Due 3.3V regulator seems
able to cone with the required current
demand. As supplied, the TFT display has
{wo options for controling the backlight
brightness and these can be selected by
bridging appropriate pais of pads on the
display board. However, | have only tried
the option in which the backlight is fully on
all the time.
‘SOFTWARE. Arduino programming is
facilitated by the fact that there are many
tutorials and code examples that can be
ddownlcaded from the web, together with
code libraries to door interface with almost
anything. Inthe interest of brevity, I will
discuss in detall only a few of the more
important parts ofthe panadapter code, but
the full isting contains many comments
to-enable a fuller understanding for those
readers wishing to make mosifications, The
full code listing is available from me
Since the | and Q
data samples are each
represented by a 12 bit
‘wor (specifying one of
are then converted into
‘two numbers each lying
within the range 21; thus
the polarity of the original
sampled AC waveforms is
preserved.
Although this code
fragment looks simple,
‘two further factors need to
be considered and these
are to do with the rate at
‘hich the incoming | and
data is sampled, since according to the
Nyquist criterion, a signal with a bandwidth
(of B Hz needs to be sampled at arate of at
least 2B samples/sec. Thus the maximum
signal bandwiath that can be displayed by
the panadapter will depend on how fast the
Due's ADC can acquire samples. There Is
also another requirement for fast sampling
‘and this isto do with the fact thatthe I and
Q data are not sampled simultaneously
but rather sequentially since only a single
ADC is used, together with a multiplexer.
‘The resulting time delay between sampling
first the I channel and then the Q channel
will give rise to a phase error for the Q
Cchennel sample and this, together with
‘any unbalance in the amplitude of the |
‘and Q channel samples will give rise to an
unwanted mirtor image on the spectrum
display, The code fragment in Listing 1
tries to minimise the phase eror by frst
sampling the land Q channels and then
doing any processing that is requited before
saving the data inthe array (@datal J,
Tne second way in which the phase error
can be reduced is by fine tuning certain,
Constants that control the way the ADC
operates. By default, a single anafogRead)
2).
4096 voltage levels), these
‘command takes about 39us to implement
but by manipulating the CortecM3 ADC
MR (Mode Register) constants in software,
this time can be reduced to about 4us. The
‘code to do tis is shown in Listing 2.
Further details on this technique can
be found in the Atmel SAM3XBE ARM
Cortex-MI3 CPU datasheet (4.
The time taken to acquire the desired
number of samples of I data and Q data
Is noted and from this an estimate of the
‘equivalent sampling frequency can be
calculated. This fs then trimmed to the
required figure using additional delay
inserted into the sampling loop using the
‘commend delayMicraseconds (Delay). The
integer number Delay associated with this
‘command must be determined empirically
but, 2s a guide, a value of 65 in my code
‘gave an effective sampling frequency of
13kH2 (equivalent to a displayed spectrum
of ++5kHz) and a value of 4 gave a
sampling frequency of 62kHz (equivalent to
a displayed spectrum of +:24kH2). It will be
noted that the displayed spectrum width is
less than might be expected from the quoted
sampling frequencies. The reason for this
willbe discussed later
Before further discussion of the software,
we fist need to consider what information
will be displayed to the user. | chose to
make both a spectrum display and a
‘waterfall available simultaneously, but
it would be a simple matter to ater the
software to choose either one or the other.
Irrespective of this, we should lm to utilise
the TFT display pixels as effectively as
possible. Hence I chose to use the display
in landscape mode, ie the display axis
‘corresponding to 800 pixels was used for
the frequency axis. The TFT display can
display up to 480 pixels of infomation
along the other ais, o I chose to use 200
‘of these forthe spectrum display amplitude
axis, 200 for the waterfall display time axis
and the remaining 80 for axis labels
(on request via e-mail
The fist part of the code is
concerned with sampling the
incoming low frequency | and Q data
from the receiver. This data will be
in the form of two AC wavetorms but
the ADC inputs on the Due board can
only deal with uri-rectional signals
The Solution is thus to DC bias the
AADC input pins at their midway point
of 1.65V via a potential divider and
to feed in the | or @ signal via a DC
blocking capacitor. By default, the
Due's ADC is configured to output 10
bits of data to represent each data
sample, but this can be extended to
12 bits using the command
analogReadResolution (12);
‘Samples ofthe incoming | and @
data are acquired using the code loop
shown in Listing 1.
ae phy
By -oeaue
FIGURE 2:1 and @ channel amps
and soft keys.
‘A decision now has to be made
about how many | and Q samples,
are to be processed each time 2
frequency spectrum or waterfall line
Is displayed. The | and Q samples
represent snapshos of voltage
waveforms in the ime domain and
these need to be transforned into
information about the amplitudes of
the individual frequency components
‘making up the spectrum display. The
‘most efficient way of doing this is by
sing the fast Fourier transform (FFT)
but this can only deal with data sets
made up of 2" pais of numbers. N is
an integer, so for example N= 10, a
data set would consist of 1024 | data
samples and 1024 Q data samples,
thus giving rise to 1024 frequency
components. The TFT display, 26
Technical Feature
Sheffield. ac. uk
FIGURE 3: Generic low pass fier (see Table 1)
however, can only display 800 frequency
Components ata time so we either have
[Link] a value of N=9 (equivalent
to 512 frequency points) or N= 10. In
the first case, not all the available pels
(on the display will be used (512 out of
'800) whereas in the second case, we
can only display 800 out ofthe available
1024 frequency components. There are
advantages in choosing the second of these
alternatives; firstly, by using 1024 rather
than 512 points in the FFT algorithm,
the frequency resolution of the displayed
spectrum is increased, even though we
cannot display the full spectrum that a
given sampling frequency provides. Also,
the displayed | and Q data is effectively
bbeing oversampled and this will ease the
task ofthe | and Q low pass filters; hence
the reason why in the earlier discussion, it
was mentioned that the displayed frequency
spectrum did not seem to match up with
the quoted sampling frequency. So for
‘example, if the nominal sampling frequency
were SDH, this would ordinarily result
in a potential spectrum display of
tumed into powers by squaring and are
normalised with respect to the maximum
signal power that could be measured by the
Due's ADC. This maximum power is that
resulting from an ADC reading of +2048.
Finally, since a logarithmic amplitude
display is required, each data point is
processed accordingly.
A simplified code to carry out these
operations is shown in Listing 3.
‘Tuming now to the TFT display, the
software for driving this is handled by a
library called UTFT that can be downloaded
from {6}. The library can dive a large
number of different displays and contains
routines for drawing text, figures, lines,
Circles ete as well as full control of colour
{and font. Since the TFT display also has an
integrated touch screen, provision has been
made to incorporate a numberof soft keys
to-contret such things as displayed spectrum
width and display amplitude range. The
handling ofthe soft heys is controlled by
another library called UTouch that can also
be downloaded irom [6]
SOFTWARE REFINEMENTS. Ifthe software
and hardware are used as described thus ‘ar,
‘wo artefacts will be evidert on the display
Firstly, a large spike will aways be seen at
the centre of the display and this corresponds
toa signal a DC. This arises from the fact
thatthe | and Q waveform voltages will most
probably have some associated DC shift
‘associated with them, If this is not removed
in the data processing then it will show as 2
central frequency spike.
‘The other artefact on the display will
‘manifest itself as a mirtor image of a wanted
spectral component, which appears in the
‘other half ofthe display. So for example,
ifa real frequency component appears at
frequency +f in the right hand half of the
displey, an unwanted image of this may
‘appear at ~f inthe left hand side of the
display, or vice versa. This effect is caused
by amplitude and phase errors occurring
in an | and Q data pair. The amplitude
‘error arises when the gains of the | and @
signal paths are unequal; the phase erors
aise because the | and Q data channels
are not sampled simultaneously but
:25kHz (the left and right hand sides
sequentially: Both the amplitude
ofthe spectrum are independent since eH and phase errors will vary with the
‘we have sampled both the | and Q frequency being displayed so a
data at BOKH2). In practice, the actual | 'ewegamp Lety ay gS g IehelADC | correction at one frequency will not
displayed spectrum width would then = = hold exactly at another. There are
be (800/1024) x 50kHz ~ 39kH2. nw schemes that can provide correction
Having finaly decided on the at all displayed frequencies but these
number of| and Q data sampies to be take up far more computing resources
acquired, 1024, these are assembled a than we have at our disposal; hence
into an array of complex numbers | adapted 2 correction scheme that
ofthe form real part=/Qdatal0I, admgiane Lott r29] aammniac | uses the received spectral component
imagiascy part-chistet, rel Hgy gaa Fee with the largest amplitude at any given
part=IQdatal2}, etc. This array is ™ " time asa test signal (71. The suing
then processed using the FFT to ew amplitude and phase corrections to
find the frequency spectrum of the the displayed spectrum and waterfall
sampled | and Q time waveforms. The sav are thus oniy valid, in theory, ata
resulting data is an array of complex
voltage amplitudes for each frequency
Point in the spectrum. Conveniently,
the Arduino IDE for the Due board
contains a set of DSP algorithms
specially writen for the Due's ARM
Cortex M3 processor, (51, and $0
these are utilised wherever possible in
the panadapter code. Further details,
are given in the first sidebar. The
FIGURE 4: Relay diver for switching low pass fers.
single frequency but in practice the
results are quite acceptable over
the full displayed frequency range.
Furthermore, the corections are made
automatically and are dynamic, ie they
change from one data set to the next.
Because the mathematics of the error
cortection process is rather involved, it
will not be discussed further here but
Interested readers will find fuller details
complex voltage amplitudes are then
inthe box on page 30. IMPLEMENTATION. A breadboard
prototype ofthe complete panadapter is
shown in Photo 1. The | and Q chennel
amplifrs, low pass filters and relay
‘switching circuits were all built on single-
Sided printed circuit boards using wire-
tended components. As can be seen from
the photo, the resulting boards are rather
large and would benefit from a redesign
using SMD parts.
‘The panadapter needs voltage supplies
(of 8V, 3.3V and 2.5V, The frst two of
these can be oblained from the Arduino
board itself and the third is provided
from the 5V supply via a simple potential
divider, as shown in Figure 2.
Photo 2 shows a close up view of
the TFT screen when the panadapler
was connected to my KX3 and the
latter was being used to monitor signal
activity around L0.14MHE. In this case,
the bandwidth of the dispiay was set at
:5kHz. From left to right, the soft keys
atthe top ofthe display can be used
to change the amplitude range of the
displayed spectrum and waterfall displays
and the displayed bandwidth (toggled
Dbetwoen :1.2kHz and 4:5kH2). The
Curent value of the former is displayed to
the left ofthe soft keys
CONCLUSIONS. When | started this
project, | totaly underestimated the time
and effort it would consume but the end
result has made it worthwhile. Although
not yet boxed-up, the panedaptr is
already proving to be a useful addition
to my KX3. Possible modifications might
include increasing or decreasing the
displayed frequency bandwidth, adding
‘mote soft keys for inceased functionality
oF utilisation of the SD card slot on the
TFT display. The latter would then enable
the panadapter to be used as a beacon
‘monitor and data logger for propagation
studies.
‘TABLE 1: Component values for low-pass fiters (all other values are common to
both filter types)
Component Namow band 4kHz Wide band 20kHz
RO, R3, R7, RB, RI2,R13,R17, RIB BK ak
(62, €8, 65, 05, 8, €3, C11, C12 4n7 1n0
‘TABLE 2: Connections between the Arduino Due board and the TFT display.
TFT pin TFT function ‘Arduino Due pin
1 ov
2 [Link]
3
4 RS 38.
5 wR 39)
6 RD Connect 3.3
7 DBs 22
8 59 23.
9 B10 2a
10 ett 25,
1 B12 26.
2 Bis: 27
13 B14 28.
4 B15: 29.
15 cs Dao.
1s
7 RST pat
38
19 LED A Connect to 3.3, Link ‘always on’ pads on back side of TFT module
20
2. 80 37
22 Bl 36:
23 B2 35,
24 es. 34
25 Dea 33.
25 DBS 32.
27 86 31
28 87, 30.
28 Tok D6
30 Tes Ds
31 TDN Da
32
33 1 DOUT D3
34 IRQ be
Technical Feature
WeBseanch
U1 Roeky SOR software vdeo comfochy
[2LATiay Python Panaptes, Martin Ewing, ABE,
sr Api 2018
[31 Act Low Pats Fiter Bese, Toasters
wr comtSt0%049
[Al Atmel SAMSXBE ARM Corer MS CPU datasheet
orto conmagesdoc1 1057p
ISI CMSIS DS trary fr ARM Cortex M3 processor
worl comsekeeMSISOSPimlinge Nin
[61 UTFT and UTovch ates
itp [Link] phd 52
(U1 eorecion Churchill FE OgarGW an
Thompson B J, The coneten of ard Q ars in
‘a cabret proses, IEEE Tan Aaspce and
ectone Systems, ES 17, No, la 198,
post 137
[61 Enabing tte oun Gatex M3 OSP functors
np tum arulnacendox prprPHPSESSIO=taesiol
‘elon aour phic Atope=140107.0
Enabling the Cortex M3 DSP functions
The folowing information has ben taken
from
1. Find the tet le platform. tt at
CulProgram Files (85)/Arduinonardware/
‘duinofsamiplatfrm.t
2. Using Notepad, amend the fle section
called
‘#4 Combine ge sections, archives, and
objects
0 Itnow includes a reference to fle called
Liborm_cortexM3l_math.2
‘## Combine ge sections, archives, and
objects
‘[Link]="{compiler
path }{[Link]}” {[Link]
flags) mepu={[Link]} * T{buil.
variant path) {buldJescript}" * Wl,
Mop {[Link]/(ouité. project name).
‘map” 0 “{build,path) {build project_
rname}.elf” * L{build pathy” Im lgce
‘mthumb W, cref Wl, check sections
Wi, gc sections Wi, entry=Reset_Hon
ler Wl, unresolved symbols=report all
WI, warn commen WI, warm section
align Wh, war unresolved symbols
WI, start group “(build pathsyscals_
‘sam3.c.0" {object fies) “(ouildvariant.
path {ould variant system ib)" "(bile
variant path) libarm_cortexM3/_matha”
{build path\farchive_fle}” WI, end
aroun
3, Navigate tothe IDE folder
Ci/Program Files (#86)/Arduinonardware/
_ardunjsam/system/CMSIS/OMSIS/Lby
ccc
In that folder, you should finda le called
Iibarm _cortexM3! math 2
Make a copy ofthis fie and place itn the
IDE folder
CulProgram Files (85)/Arduinolhardware/
_arduinoysam/rariantsfarduino_due_x!
The Cortex M3 DSP functions should now
be accessible,
27 30
Technical Feature
Conection oft and @ errors
This escussion isan abridged version of that found in (71
[A simplified and ero ree version ofthe | and @ data being passed
Into the panadapter can be writen as
I(t) =A cost and Q(t) = Asin ot
This pair of signals canbe thought of esa complex sinal
1) +jQ@ = Aelot
In proce the recive | and @ data wil contain eros such that
©) = (1. +e)A cost +a and Q,(¢) =Asin@we-+ 9) +b
where
€ isthe fractional amplitude imbalance
isthe phase imbalance
2 isthe DC ofset in the I channe., and
is the OC offset in the @ chanel
1 and b can be coecte by subtracting the average level from the signal
in each channel
Then we have
I(2) = (1 +e)A cosaat and Qp(¢) = Asin(ot +9)
The | and Q, signals ae treated as vectors and two comection coefcents
P and E, are required, P for rotating one vetor and E, fr scaling the
oer Then the conected signa |, and Q, are related to, and Q, by
I= Eyl and Qs =P hh + Q
Itcan be shown thatthe required form for E, and Ps
cosy
A tee)
=n
ae
ter application ofthe correction proces, the fnal signals are
BO =
‘An amplitue scaling factor ens has been introduced but since tis
common to bath |, and Q, this is nat af importance,
cospcoswt and Q3(¢) = Acosg sinwt
The procature to obtain the conection cosets isa follows:
The | and Q signals are each sampled 1024 times ard the dat
stored in aray IQdata1[J. The 2048 data values are also store in
ray IQdata2] fo future use. The data in JQdata is next FFTed as
Outined in the previous discussion ofthe panadaptersotware. Now
the fQdata11 array contains 1024 complex frequency values arranged
in pairs (eal part, imaginary pat, eal part, imaginary pat...) inary
locations (01 to 12047,
‘The DC ofse's i the orignal | and Q sampled time waveforms re
obtained from
Ledata10} ang p= #2éetany
‘and these values wiE later be subtracted from all he data contained in
aay IQdat2217.
To obiain the comecton coecintsE, and P we need to locate the
complex equency component in array fAdataI( whic has the largest
‘magrituce and to note is location within the array as a par of indies,
eg lQdated(2*1] real part of fequency, Qdstal (24/+4 imaginary
pa of frequency. This frequency component can be thought of as our
teat signal and let us assume that te values of ts eal and imaginary
parts aro
IQdatal(2 + i] = W and 1Qdatal|2 +i + 1]
‘Thon the unwanted image ofthis test signal wil be located within the
array 1ddatatL) at
adata112048 2*i) 10a pat of image frequency ana
qdata1£2048 2414-1) imaginary part of image frequency.
‘Theo associated data values are
1Qdatai [2048 ~ 2 i] = ¥ and 1Qdatai[2048— 2+ +1) =z
Itcen be shown that te corection coefficients E, and P are related 10 W,
X,Y and Z by
demom = 2/(W-41)%W +) +0 210K 2);
ol = 1. YHW+Y) 240 Z)4donom;
Bm ZHWHI+Y"0 2))*denom;
Finally, we can apply these correction factors to the copy ofthe eriginal
sampled | and Q data which was previously stored in array data?
111Q correction to copy oferiginal sampled and Q data
for (uinti6_t1 = 0; 1 < ftSize; ++)
{
{/1emove DC shit
/Qct0212" 2;
This data is now FFTed to obtain the coracted frequency spectum which
will subsequently be displayed,
Listing 1: Simple code loop to sample | and REG_ADC_MR = (REG_ADC_MR & to implement FFT *)
Qdata ‘OxFFFOFFFF) | 0100020000; arm_efflradixa_£32(8S, 1Qdata);
Const uint32_t ftize = 1024 change PRESCAL from 2 to 1 1° Process the data through the Complex
const float weight = 1. /2047.; ‘REG _ADCMR = (REG_ADC_MR & Magnitude Module for
fofuintt6.t ‘OxFFFFFOFF) | 0100000100; cafeulatig the magnitude at each bin.
FRetum data in array Spectrum; This
flootrew! 2047) * weight;
1Qdatal2%i+ 1) = foatrawQ 2047) “weight;
¥
Listing 2: Reducing the ADC read time.
i modify 3DC_MR register to make ADC read
faster
{change STARTUP from 8 to 2 (S12 to 15
‘periods of ADC clock)
Listing 3: Converting the sampled | and Q
‘data toa frequency spectrum,
const uint32_¢ 1tSize = 1024;
‘const uint32_tifFlag = 0;
const uint32_t doBitReverse = 1;
(Unitiaize the CFFTICIFFT module
‘arm ct radhx4 init 132188, fice, ifFlag,
oBltReverse);
1° Process the complex 10 data in array
‘1Qdatal] through the CFFT/CIFFT module
arm_emplx_mag_f32(IQdata, Spectrum,
1" Normalise the frequency data and convert
for (uint 16. i=0; f
