-
-
Notifications
You must be signed in to change notification settings - Fork 16
Reference
DDS waveform generation (we might be interested in Figure 6 operation)
Possibly a really good reference book for this project w.r.t. ionosphere: http://www.agentschaptelecom.nl/sites/default/files/near_vertical_incidence_skywave_-_interaction_of_antenna_and_propagation_mechanism.pdf
- Sec 2.2.5, 2.4 NVIS Propagation modeling
- Fig 2.7: NVIS amplitude fading (in any communications model, there is an amplitude fading factor vs. time like this)
- Sec. 2.5: references to multiple kbps NVIS channels using narrowband modulation
- Sec 5.3: NVIS antenna performance
- Chap. 8: advantages of using 2 NVIS antennas and 2 simultaneous receivers (such as Red Pitaya has) ~ 10 dB Fig. 8.10
- It appears they were interested in 16kbps in a 3kHz channel (most modems do perhaps 300-1200 bps in such a channel)
Masters Thesis 2012 using NVIS in Middle East (Fig 3.6 shows stations) -- gives a lot of general info, for your background reference. https://www.scbaghdad.edu.iq/library/Astronomy/MS.C/2012/Studying%20MUF%20&%20FOT%20Parameter%20Variations.pdf
Measurements in Scandanavia using narrowband 75bps - 2400bps STANAG 4285 modems -- reference to what's possible for emergency communications using classical techniques http://onlinelibrary.wiley.com/doi/10.1029/2000RS002547/pdf
The problems these systems have in common for our experiments and deployments include]
- too expensive radios ~$1000 ham radio transceivers
- too little instantaneous bandwidth: < 3.5 kHz typically.
These systems fail to use Shannon-Hartley theorem optimally. One can trade bandwidth and/or time for power<-->SNR. In essence we are proposing a more optimal use of scarce RF resources by using low-power pseudonoise broadband waveforms that other users will scarcely detect.
Numerous other crowd-sourced radio science projects out there, including:
- Radio Jove 20 MHz receiver network listening to Jupiter's decametric radiation. Seems to use spectrum vs. time.
- Aurorasaurus actually for optical measurements, but has great citizen use and visualization.
Reference for Raspberry Pi that we didn't use
example setup and coderpitx code is capable of IQ modulation (amplitude and phase modulation real & imaginary), via PWM. Even if we don't use the Rpi PLL, maybe this code provides clues as to IQ modulation via PWM.
Perhaps a $20 RTL2832-based USB receiver can do the job. SpyVerter costs $49 and is said to be of good quality (but good enough for this project w.r.t. jitter?). If the receiver doesn't have the frequency range you want to listen in, you can use an upconverter to get in range (e.g. filter and mixer).
This may be implemented as a fixed-frequency oscillator, bandpass filter, and mixer. Here is an integrated RTL2832 with convertor etc. for under $50 covering 3 kHz to 1.7 GHz. You can find several more integrated and modular frequency convertor solutions.
One of the first things to tackle is whether it's wiser to just use a PLL or DDS chip that can talk to any singleboard computer, perhaps using an eval kit. Although it does seem cool to use the built-in Raspberry Pi PLL, there are also upsides to not being dependent on one piece of hardware that could change at any time.
Synthesizers choices would be along the lines of:
- AD9834 $10 chip up to 75 MHz center frequency, DDS arbitrary modulation, AD9834 eval board under $100.
We will need to consider which parameters we need to modulate (amplitude, phase, frequency) with our psuedonoise waveform.
A question with using Rpi or Beaglebone with SDR is does it have enough CPU to do a loopback test?
Project for Raspberry Pi running HackRF -- but could an Rpi3 run 2 HackRFs simultaneously or one HackRF and one RTL2832 simultaneously?
Descripton of BeagleBone Black with Ettus B200 -- the Raspberry Pi 2 and 3 both generally have a lot more CPU power than the BBB. But there are some places w.r.t. I/O that the BBB still wins. Not sure if they apply here.
| proc | rx | tx | cost |
|---|---|---|---|
| Rpi 3 | RTL2382+ 2x upconverter | AD9834 DDS + MPU | 35 + 2*50 + (10+40) = $185 |
| Rpi 3 | Ettus B200 | (same B200) | 35 + 715 + 0 = $750 |
| NUC / i5 | RTL2382 + 2x upconverter | HackRF One | 520 + 2*50 + 324 = $944 |
| NUC / i5 | Ettus B200 | (same B200) | 520 + 715 + 0 = $1235 |
These measurements and cheaper price of Red Pitaya led us to use much higher performing Red Pitaya to handle RF.
Initial experiments show that the CPU & PLL on the Raspberry Pi 3 may be capable of transmitting HF radar waveforms with sufficient spatial resolution to resolve interesting atmospheric disturbances. Implicit in that is the ability for the Raspberry Pi 3 to transmit broadband data over the horizon on HF using the ionosphere as a giant natural mirror 200 km in altitude.
Initial testing showed 20 ms jitter on transmit waveforms with Raspberry Pi PLL--need to reduce this by at least one order of magnitude to get useful radar resolution.
Command used to test jitter.
We tweaked the Raspberry Pi's /boot/config.txt as per link above to help jitter.
sudo rpitx -i tx_code-l10-b100-000000.bin -f 50000 -s 100000 -l -m IQFLOAT -p -53
Jitter was measured with a USRP N-series locked to GPS. I believe we can borrow an equivalent system to measure jitter, as it would otherwise cost a couple thousand dollars for this reference equipment. In the meantime, we can simply use a cheap receiver--if we don't see jitter, there must be little jitter. If we see jitter, then we can use the expensive setup to see what the jitter source is (TX or RX).
For loading/decoding the waveform you can use the standard filesink in gnuradio.
You can also use the digital_rf filesink.
Much geoscience radar community radar processing is nowadays done using digital_rf, so that should probably be the format to aim for in digitizing raw voltage data.
It shouldn't be too difficult to create digital_rf from a rtl_sdr dongle instead of the USRP.
We generate a repeating pseurandom binary phase modulated waveform, 1000 bits and 100 microsecond baud length
We test in lab doing an RF loop back. Send a short coded sequence and record that.
Juha's initial look seems to show frequencies are stable on the long term. They are approximately as accurate as the quartz crystal on the Raspberry Pi. The timing of phase code edges has a 20 microsecond / 1 ms timing error standard deviation, which is probably related with granularity of timing in software.
We don't think of these systems as competing but rather complementing. Other types of instruments e.g. optical are also experiencing the revolution of dense networks of small cheap sensors completings large single site sensors.
- SuperDARN: excellent global network running for over 20 years, costing just over $1 million per site. SuperDARN observes on timescale of minutes in the 8-20 MHz range. We propose time resolution on timescale of seconds in the 3-10 MHz range.
- Arecibo this is a UHF incoherent scatter radar studying the ionosphere from a single site and with mechanical steering. Naturally, a 330 meter dish antenna in the mountains is very expensive.
- FAST and GBT similar: cost/size/single site
- HackRF One $300, 20 MHz bandwidth. HackRF One block diagram
- BladeRF x40 $420, 40 MHz bandwidth. BladeRF block diagram
- Ettus B20x $750, 56 MHz bandwidth. Several of the Ettus radios are sophisticated implementations of an AD9364 ecosystem.
The above are all good to excellent choices for their purposes, which often is on-air cellular, Wifi, bluetooth or high-resolution radar. However:
- Inexpensive <$100 PCs currently have trouble handling the full bandwidth of these radios.
- we need < 1 MHz bandwidth
These radios take the approach of using a frequency synthesizer and IQ mixer for highly frequency-agile (several GHz) broad bandwidth (tens of MHz) transceivers.
By distinction, our application requires only several MHz of frequency agility and < 1 MHz bandwidth. It is possible that we can omit the IQ mixer and external DACs and simply modulate the frequency synthesizer itself, saving a great deal of cost and complexity.
To do so probably requires a microprocessor or FPGA to clock in the pre-programmed modulation sequence. The AD9834 eval board has a Blackfin MPU enabling control and modulation if we used the Raspberry Pi -- but the Red Pitaya is cheaper than using Rpi and has much better specs.
Normal COTS amateur radio gear struggles to pass more than about 2.5 kHz instantaneous bandwidth with reasonable amplitude flatness.
- most nations allow at least 6-10 kHz instantaneous bandwidth for HF amateur radio transmissions, and bandwidth ~ range resolution, so we want the maximum legal bandwidth -- and may employ experimental licenses such as MIT Haystack currently holds to use more than 10 kHz bandwidth.
- due to multiband coverage from 160-10m and beyond, ham radio oriented SDRs cost >$600 and require a desktop PC to operate.
- a partial SDR transceiver is the Pocket SDR for $500 early adopter price, but it is a test device not yet ready for general use according to the creator. We could see if the pricing will change for a future run.
We may need to avoid burning out the front end of the receiver on transmit. A PIN diode or mechanical reed relay can be used for this purpose. Given the low power and frequent switching, a PIN diode may be the suitable choice.
example of PIN diode switching You don't need the specific filters they used, it was just an example application.
There are caveats of using PIN diodes at low frequencies, however. Consider we may want to operate as low as 3.5 MHz. Carrier lifetime is the key specification for PIN diodes, you'd like to have it be several cycles at your lowest frequency. E.g. 2 MHz <-> 0.5 μs cycle time. However, I have seen amateur designs such the ARRL one above using non-PIN diodes like the 1N914 or even 1N4001 at these frequencies. Those cheap diodes may be a future possibility but for now let's consider starting with the "good" $3 PIN diodes. An example PIN diode covering 100 kHz to 1.5 GHz is Macom MA4P7102F for $3 each.