Simple and Accurate Variable Frequency RF Signal Generator

Transcription

1 Elwood Downey, WBØOEW 8274 North Sunset Ranch Loop, Tucson, AZ 85743; Simple and Accurate Variable Frequency RF Signal Generator This Generator uses an Arduino-controlled direct digital synthesizer stabilized with a GPS receiver, along with a hand ull o parts. This generator produces any requency between 5 khz and 4 MHz with an accuracy approaching one part in 9, or example. Hz at MHz. It uses an Arduino Nano, a GPS receiver with antenna, a digital encoder, a small TFT LCD color display, and the Silicon Labs Si535A direct digital synthesizer (DDS). Together these parts cost me about $5. I will also share some interesting applications or this device. Operation My housing and user interace are shown in Figure. The user s desired requency appears across the top o the display. Tapping the screen will underline dierent groups o three digits. Rotating the encoder will change this group, with aster rotation changing the numbers even more rapidly. The next row reports the error measured during the previous second measurement scaled to the Figure Packaging and user interace. Figure 2 An inside view shows the method o construction. QEX September/October 28 3

2 current user requency and includes quantization errors rom loading the integral DDS registers. This row can also report i the GPS PPS (pulse per second) signal is not available or the reerence clock is not working. The next row reports the approximate output power in dbm into 5 W. Clicking the encoder cycles through our power levels. The remainder o the display shows the current date and time, antenna location, an allsky map showing the tracks o the satellites currently being used or the ix, the sotware version, and the best and worst satellite signal-to-noise ratio. These extra data are just or un and are not used by the generator control algorithm. Note that displaying the time is given low priority and can be as much as one second late. Construction Figure 2 shows an inside view o my unit showing my method o construction. The modules are placed on one piece o per board and wiring is point-to-point underneath. The display is stacked below in this view. The wiring diagram is shown in Figure 3. The level shiting diodes ensure the Arduino is reliably triggered rom the 3.3 V outputs rom the GPS PPS and Silicon Labs DDS. I used the Arduino Nano but any Arduino should work with suitable attention to pin numbers. The unit can be powered with USB but I wanted to use my shack 3.8 V dc supply, so I added a USB buck converter rom ebay to create. For the Si535A I tried both the Etherkit 2 breakout with the temperature controlled crystal oscillator (TCXO) option and the Adaruit board 3, which has no temperature control. Exposed on the bench, the unit with oven control was signiicantly more stable, but inside the enclosure I could measure no dierence, so the inal choice just depends on your packaging preerence. See Table or a list o the main part numbers and current prices in US$, not including wires, connectors and enclosure. I did not use any third-party sotware libraries or the Si535A, preerring to write code or the bare metal to allow me to develop a deeper understanding o how best to use this synthesizer. Similarly, I did not use the Adaruit GPS library but only because I needed just a ew eatures so I could save precious memory by writing my own code. I did use the Adaruit GFX, ILI934 and FT626 libraries or their TFT display and touch overlay panel. My source code is available on the web page. You may also check or updates or this project at 5V J7 Etherkit Arduino Nano TFT J8 SCL SDA D5 A5 A4 D9 D D D3 D2 D3 D4 D8 D7 D6 CLK MOSI N Ω CS D/C SDA SCL 56 Ω B A S N448 Encoder PPS TX RX Figure 3 Wiring diagram o the generator. QX89-Downey3 C Ant GPS 4 QEX September/October 28

3 Table Parts list not including enclosure and various wiring materials Part Cost Adaruit Ultimate GPS breakout PID 746 $4 Adaruit GPS Antenna External Active PID 96 $3 Adaruit ufl to SMA adaptor cable PID 85 $4 Adaruit 2.8 TFT LCD with capacitive touch PID 29 $4 Adaruit Rotary encoder PID 377 $4.5 Either Adaruit Si535A breakout PID 245 $8 Or Etherkit Si535A breakout with TXCO $6 Arduino Nano several sources $22 N448 diode, 2, $.5 56 W ¼ W resistor, 2, $.5 WBOEW Timer Counting CLK up p/ QX89-Downey4 d/2 QX89-Downey5 icr_ CLK = Pulses Per Second n_overlows = 4, m_secs = 2 icr_ icr_ * (n_overlows ) + icr_ m_secs Figure 4 An example measurement sequence. h d/2 p/2 WWV Figure 5 Idealized path geometry between WBØOEW and WWV. Theory o Operation I the GPS receiver can receive rom at least our satellites it produces one digital pulse per second (PPS). Clock o the Si535A is programmed to produce a requency o 5,,. Hz. This is ed to an Arduino counter 4 whose value is captured at each PPS signal. These measured counts are used to correct the Clock requency. The claimed accuracy o the PPS is one part in 8 with random jitter. This could be improved another actor o by applying the error measured over seconds, but I did not (Hz) want to assume the DDS would be perectly stable or this long, so instead I measure the error over a period o seconds and apply one tenth o this as the correction. The sotware continuously compensates or the DDS crystal oscillator error by measuring the value o a known reerence requency, REF_FREQ, being generated with CLK. It then uses this correction to produce any requency on CLK to the same relative accuracy. Figure 4 shows an example measurement sequence. The DDS CLK is set to generate the REF_FREQ. This is connected to D5 (T) so it causes Timer to count continuously rom to then overlow and repeat. Each overlow triggers an interrupt that increments n_overlows. Meanwhile, the GPS PPS is connected to D8 (ICP). Each alling edge copies Timer (TCNT) to the Input Capture Register (ICR) and triggers an interrupt. The irst PPS interrupt ater a measurement is started saves ICR in icr_. Ater m_secs more PPS interrupts the ICR is saved in icr_. Thus the total counts during m_secs is the remaining Timer counts until the irst overlow, plus the number o counts in the next ull n_overlows periods, plus the counts until the inal PPS. The Si535A uses several ixed point coniguration registers o the orm a + b/c. Care is taken in the sotware to choose the best combination o these integral values to achieve a total precision o at least 3 bits or about one part in 9 to match the eective GPS precision. The desired user requency is generated on Clock rom the same crystal time base so it is also accurate to the same precision, scaled proportional to requency. The reerence corrections run continuously in the background so the user is ree to adjust the desired output requency with the digital encoder on Clock at any time. Note well that this DDS generates square waves, not sine waves. This is ideal or the control purpose on Clock, but the Clock user output will be rich in odd harmonics. Entering the ARRL Frequency Measurement Test I tested the project by using it or the ARRL Frequency Measuring Test held in April 28. This test required measurement o requencies in the 2, 4 and 8 m bands. I tied or second place in a ield o 8 with a maximum error o.9 Hz. I used an Elecrat KX3 with SpectrumLab 5 running under wine on an imac with macos.3.3. This is an exceptionally versatile and accurate audio measurement tool built or Windows, but I had no problems running it on macos with wine and I have no reason to doubt it would run equally well under linux. The File Export Format dialog contained the ollowing threecolumn deinitions; note the third column records the time in my local time zone o UTC 7 hours: # WWV eak_(99,)###.# # GPS eak_(965,975) ###.# # UNIX time-252#########. I did not make a direct connection between the generator and the receiver, I just used a -inch bare wire lying on the desk top near the receiver with one end attached to the Clock output. I set the receiver to USB and tuned so I could hear the FMT signal at about 5 Hz audio tone. I then adjusted the generator and the wire to produce a second tone at about 45 Hz with about equal intensity. I then recorded the audio with both requencies. Later I measured each audio tone peak requency with an FFT bin size o 2 mhz. The requency o the carrier was then computed as the dierence between the audio tones added to my generator requency. To ind each FMT requency I applied the ormula, QEX September/October 28 5

4 Figure 6 Doppler shit and ionospheric height change over a 24 hour period. = + ( A A ) FMT GPS FMT GPS where FMT is the unknown FMT requency, GPS is the requency displayed on the GPS generator, A GPS is the audio requency o the GPS generator reported by SpectrumLab peak_(), and A FMT is the audio requency o the FMT signal, reported by SpectrumLab peak_() Note that by using the dierence in the measured audio tones this technique relies entirely on the GPS generator or accuracy. The Arduino, receiver and computer time bases must be stable during the test but need not be well calibrated. Computing Change in Ionospheric Height by Measuring Doppler Shit I tried measuring the Doppler shit o WWV at 5 MHz. This allowed me to calculate the change in path length rom their transmitter in Colorado to my station in Arizona. The distance is about km so I assumed the path required one ionospheric relection and no ground relections. Using this model I estimated the eective height o the ionosphere over time as ollows. The velocity, v(t), at which the source o an electromagnetic wave is moving with respect to the receiver can be computed rom its Doppler shit as a unction o time 6, D() t vt () = c () In my case, the source is not moving but the path length is changing because the eective relection height is changed by insolation. For example, i the requency is increasing, it means the path length is getting shorter because the ionization layer is getting thicker and the eective relection height is getting lower. Integrating velocity rom time to time T lets us ind the total path length change, Dp, in this interval as, c T D p( T ) = () t dt D c Dt T Dt i= D i (2) This integration can be perormed numerically by maintaining a running sum o each measured requency value with respect to the known reerence requency every Dt seconds. To convert this path length change to a change in height, I assumed the path simply ollows each hypotenuse o two right triangles sharing a common vertical side beneath the point o relection as shown in Figure 5. From this geometry we have, d p = h (3) then rom h/ p we ind the change in height or a given change in path length is approximately p D h= D p (4) 4h Combining Eqns. (2) and (4) we ind the height change as a unction o time, T Dt Dtpc D ht ( ) = D (5) i 4h i= I set h to the F-layer height 7 o 3 km, and d to km or which p is7 km. The measurements were taken every Dt = 5 s, and = 5 MHz. This yields a scale actor o 292 m s per sample. Putting it all together we have one result over a 24-hour period shown in Figure 6. This was measured using an active receiving loop antenna 8 connected to an RFSpace Cloud-IQ SDR receiver (see: rspace.com/ RFSPACE/CloudIQ.html) read with gqrx (see: gqrx.com). and piped into SpectrumLab running under wine all on macos.3.3. The radio was set to about 4999 khz USB so the WWV carrier produced an approximately khz audio tone. The GPS reerence was set to 4,999,97 Hz to produce an approximately 97 Hz audio tone. It is only the dierence in tone requencies that matters. SpectrumLab was programmed to ind the peak o each o these two requencies and store the dierence. These were summed with perl (see: and plotted with gnuplot (see: The SpectrumLab output ile was named wwv- 5-doppler.txt with three columns or the measured WWV and GPS reerence audio requencies and the local time in UNIX ormat. The key gnuplot commands include: gnuplot> set xdata time gnuplot> set timemt %s gnuplot> set ormat x %H:%M gnuplot> set y2tics gnuplot < perl -n -a -e $d=$f[]-$f[]-3; $h-=292*$d/; print \ $F[2] $h $d\n\ ; wwv-5-doppler.txt using :3 axis xy2 title Doppler shit, using :2 title Height change linewidth3 The height scale is only approximate but you can see some interesting trends on this day. The plot begins at local midnight. The height increased a bit more until sustaining a airly steady maximum or much o the remainder o the night. Ater an interesting bump an hour beore sunrise it dropped quickly then leveled o during the daytime hours. As sunset approached the height rose again quickly. The bounce back an hour later looks suspiciously like the mirror image o the bump beore sunrise. It looks as i the remainder o the night may not be as high as the previous night. These plots are easy to make and un to ponder. I have recorded several days on dierent bands in this manner and all contain varied and interesting characteristics. 6 QEX September/October 28

5 I hope you enjoy using this variable requency generator. Feel ree to contact me i you have any questions. Elwood Downey, WBØOEW, has been licensed continuously with the same call sign since 974. He enjoys building telescope control systems and related astronomical instrumentation. Elwood maintains a small Amateur Radio web page at ham. Notes 5V-2A-USB-Charger-DC-DC-Step-down- Buck-Converter-Voltmeter-Module/ ?hash=item4aec92:g:CIAA OxygPtS4cu5. 2 Etherkit Si535 breakout board with TXCO, 3 Adaruit products, tutorials and sotware are available at 4 For details on programming the counters see Chapter 2 in the Atmel data sheet at 5 SpectrumLab is available as a ree download at eect. 7 This is a typical height or the F layer responsible or HF propagation; see en.wikipedia.org/wiki/ionosphere. 8 I used this ampliier: active-antenna.eu/ ampliier-kit, with two coplanar loops m in diameter with a center height o about 3 m. QEX September/October 28 7