Full amateur HF allocation monitoring and display SDR project

[rhb]

Basic idea is to digitize and merge IQ streams for each band, present them as a waterfall display at up to 4K resolution and record all the IQ streams to USB disk if desired.  Allow collecting GPS time stamped IQ streams for geolocating rules violations such as malicious interference, obvious commercial activity, etc.

Use a GPSDO as the master clock driving an Si5332 with 8-12 outputs.  Use an FET switch based quadrature switching (aka Tayloe) mixer to convert each band allocation to baseband, anti-alias filter and digitize at 16 bit resolution using a pair of LTC2374 or similar.  Follow that by digital filtering to isolate the amateur spectrum and 50 KHz guard bands.  Generate an HDMI video stream of the spectrum.  On a 4K TV this is a visual bandwidth of about 1 KHz covering 1.8 to 29.7 MHz.  A 4 TB disk will buffer about 48 hrs of activity which can be played back at any time.

A rough estimate is that they can be built in small quantity for a BoM of $350 or less.   I have Zynq dev facilities for the digital tasks.  But if I'm to have time for that I need someone to help with the HW design so I don't have to learn a PCB design tool and can focus on just the FPGA and beyond.

Is anyone interested in collaborating on this?

[ogden]

Better just build direct sampling receiver similar to http://websdr.ewi.utwente.nl:8901/, downconvert all subbands digitally in the FPGA. Two channel 40MSPS ADC will cover 1.8 to 29.7 MHz bands, one channel running 1st Nyquist zone for 0-14 MHz, another 2nd Nyquist zone for 29.7 MHz band. When you drop 29.7 MHz support, you can get away with design which is nearly copy of mentioned receiver, just with number of DDC's in the FPGA added.

[rhb]

Much to my surprise, when I looked up the cost of a 130-180 MSa/S 16 bit ADC, it is cheaper to just grab it all and sort it out later.

I think a waterfall of DC to 38 MHz  at 10 KHz VBW on a 4K TV and the ability to zoom would be *really* cool. And I bet there is an eval board to do the heavy lifting.

[Kalvin]

There are quite many KiwiSDRs online which can display the whole HF band at once:

http://ve3sun.com/KiwiSDR/

For example this one:

http://thomas0177.dnshome.de:8073/?f=10110.0cwz6

Use the mouse scroll to zoom in and out into the band in interest.

A few KiwiSDRs could be setup to record the whole HF band  in real-time in one receiving site. Combining signals from similar stations, it would be possible to obtain the transmitter location within some range if the stations stamp the recordings with accurate GPS time stamps.

I am not quite sure whether this kind of big brother is worth the effort and resources, though.

[in3otd]

A few KiwiSDRs could be setup to record the whole HF band  in real-time in one receiving site. Combining signals from similar stations, it would be possible to obtain the transmitter location within some range if the stations stamp the recordings with accurate GPS time stamps.

it seems someone thought of that already, https://www.rtl-sdr.com/kiwisdr-tdoa-direction-finding-now-freely-available-for-public-use/

[rhb]

The KiwiSDRs are interesting.  I'd not heard of them before.  However, *very* different concept and they are much more expensive.

A broadband direct sampling system system has the disadvantage that the dynamic range is controlled by the strongest signal in the passband.    The multiple conversion scales the dynamic range to the strongest station in each band of interest.

An 8-12 channel 12 bit ADC appears to be more economical.  That should drop the BoM to under $100.

As for whether it's worth the effort, the FCC has made clear that protecting the amateur allocations from infringement by unlicensed and commercial users is primarily the responsibility of the amateur community,

And once you have witnessed deliberate malicious interference with health and safety traffic using a high power transmitter, you'll understand the need for geolocation.

However, the real motivation is the fascination of watching the waterfall and being able to replay the waterfall sped up and then be able to listen in to see what some 3 AM burst of activity was.

[ogden]

A broadband direct sampling system system has the disadvantage that the dynamic range is controlled by the strongest signal in the passband.  The multiple conversion scales the dynamic range to the strongest station in each band of interest.

It's not that bad as urban legends are telling.

An 8-12 channel 12 bit ADC appears to be more economical.  That should drop the BoM to under $100.

Please name such ADC. I struggle to find it.

I am still not convinced that 4 .. 6 preselector filters with 4..6 LNA's and 8..12 ADC channels (simultaneous >= 200KSPS each!) will be lower cost than single lowpass filter + single LNA + ADC of broadband direct sampling receiver.

[cncjerry]

Kiwi SDR works pretty well for TDOA.  I've mapped AM radio stations very precisely.  Can't really map SSB jammers unless they are transmitting a carrier or highly compressed music.  But even then, what are you going to do with the data? FCC will do nothing.  Just read about W6WBJ, self professed "worlds best jammer".  They fined him 25k, he said he wasn't paying and the DOJ decided not to prosecute.  His case has been open since 2003.  I'm disgusted but I stick with the hobby.

Jerry

[Beamin]

Basic idea is to digitize and merge IQ streams for each band, present them as a waterfall display at up to 4K resolution and record all the IQ streams to USB disk if desired.  Allow collecting GPS time stamped IQ streams for geolocating rules violations such as malicious interference, obvious commercial activity, etc.

Use a GPSDO as the master clock driving an Si5332 with 8-12 outputs.  Use an FET switch based quadrature switching (aka Tayloe) mixer to convert each band allocation to baseband, anti-alias filter and digitize at 16 bit resolution using a pair of LTC2374 or similar.  Follow that by digital filtering to isolate the amateur spectrum and 50 KHz guard bands.  Generate an HDMI video stream of the spectrum.  On a 4K TV this is a visual bandwidth of about 1 KHz covering 1.8 to 29.7 MHz.  A 4 TB disk will buffer about 48 hrs of activity which can be played back at any time.

A rough estimate is that they can be built in small quantity for a BoM of $350 or less.   I have Zynq dev facilities for the digital tasks.  But if I'm to have time for that I need someone to help with the HW design so I don't have to learn a PCB design tool and can focus on just the FPGA and beyond.

Is anyone interested in collaborating on this?

I don't know enough to work on that but definitely something that interests me and would like to see the end result.

Are you doing all that in hardware? I'm guessing this is front end stuff that can't be done in something like GNU radio?

[rhb]

A broadband direct sampling system system has the disadvantage that the dynamic range is controlled by the strongest signal in the passband.  The multiple conversion scales the dynamic range to the strongest station in each band of interest.

It's not that bad as urban legends are telling.

An 8-12 channel 12 bit ADC appears to be more economical.  That should drop the BoM to under $100.

Please name such ADC. I struggle to find it.

I am still not convinced that 4 .. 6 preselector filters with 4..6 LNA's and 8..12 ADC channels (simultaneous >= 200KSPS each!) will be lower cost than single lowpass filter + single LNA + ADC of broadband direct sampling receiver.

The relative dynamic range of an ADC is not subject to mythology.  Either you know how they work or you don't.  My RSP2 is *very* deaf when the 1 KW AM station tower two miles from me is active.  Deaf as in cannot pick up *any* other AM stations during the day when it is active. The RSP2 includes internal AM and FM BCB filters for this reason.  If the problem were merely urban myth, then SDRplay would not have added the filters to the RSP1 design.

The STM32F429 has 7.2 MSa/S of 12 bit capacity spread across as many as 24 channels.  The LPC4370 has an 80 MSa/S 12 bit ADC spread across 4 channels.The LPC4370 will handle all of 10-1.25 m and the STM32F4 will handle all the rest.  Both are available on dev boards for $20.  Most, if not all, the STM32F4 have the ADCs.  Unfortunately, the Si5332 will not go above 250-300 MHz, so one would need a different clock generator for 2 and 1.25 m to implement the Tayloe mixers for those bands.

If the KiwiSDR can locate an AM station but not a SSB station it's a software failure.  Mathematically, the type of emission is completely irrelevant.

I suspect the biggest attraction to such a monitor would be looking for band openings over the course of the previous day by way of a fast forward through the previous day's waterfall.

@Beamin.  Perhaps you should learn a bit about radio before entering a rather detailed discussion of an SDR design exercise.

[jks]

Software failure? Not so much: http://valentfx.com/vanilla/discussion/1358/in-case-there-were-any-doubts-vp6d

It's all about the autocorrelation. That particular DF took a while to get, and the error is a bit larger than usual, due to a combination of factors. An intermittent signal (contest CW), long measurement baselines (Kiwis used were in New Zealand, Hawaii and California) and of course the lids not using split.

Cheers,
John, ZL/KF6VO
KiwiSDR

[cncjerry]

It's not really a software failure in KiwiSDR, more of a limitation in the practicality of coding TDOA for modulated signals.  Think of CW dits in the time domain.  Depending on the arrival, one station might get the end of a dit and the other the space.  So one of the stations doesn't present TD data for that instant.  Constant carrier on the other hand makes this issue moot with the exception of the start and end of the transmissions.  The TDOA software needs to map the signals across each other and find the phase difference.  A human can do it manually so I guess ultimately the TDOA software will catch up, if it hasn't already.  Having written this, I have friends that would then say, "you proved my point, a software failure in KiwiSDR."

Jerry

[rhb]

If the type of modulation has any effect on your ability to do a TDOA then it *is* a software failure.  For a small number of stations, cross correlation and a linear fit to the phase vs frequency is the usual method. It is a  DSP 101 exercise: two FFTs and a linear regression for two receivers.

There are more advanced methods which require a large number of receivers, but which in turn, produce much better accuracy and resolution.

[ogden]

The relative dynamic range of an ADC is not subject to mythology.

Indeed it's not. You have problem with nearby in-band interfering station, so you make scary story out of it - to convince that everybody needs your overengineered multiband radio instead of notch filter for existing direct RF sampling SDR

[coppercone2]

this is another reason why I like the internet: its all encrypted so you don't have people doing mass surveillance.

[rhb]

The relative dynamic range of an ADC is not subject to mythology.

Indeed it's not. You have problem with nearby in-band interfering station, so you make scary story out of it - to convince that everybody needs your overengineered multiband radio instead of notch filter for existing direct RF sampling SDR

I am simply explaining the logic behind a design decision.  I want to play with quadrature mixers and this seems like a good project for that.  Curiously, at the same time that you are saying I should not use a quadrature detector, in another thread someone is arguing that one should.

[ogden]

I am simply explaining the logic behind a design decision.  I want to play with quadrature mixers and this seems like a good project for that.

Tayloe detectors are good for simple, low cost receivers that does not require top performance. Full amateur HF allocation monitoring, recording and display SDR project, especially at 350$ target BOM price, requires better receiver(s) than that. Cannot comment about LPC4370 "ADC", but STM32 have awful clock jitter, it is not fit for "350$ BOM radio application". I believe that there's no reason to build "multichannel Tayloe receiver" if for same price you can build direct RF sampling receiver.

You really shall give URL when you refer to other conversations.

[rhb]

I think the  builders of the NorCal 2030  might disagree with your  assessment.  At the moment I cannot find the PDF nor the website, but another gentleman uses a quadrature detector in an experimental 40 m CW transceiver  that achieves over 80 dB of image suppression.  IIRC the opposite image suppression is actually over 120 dB.  The transceiver is a DC design consisting of a collection of connectorized modules connected by coax so he can try out different choices on an apples to apples basis.

www.norcalqrp.org/files/NC2030/NC2030_v5.pdf

The Softrock radios have done a lot to popularize quadrature mixers, but they are low cost designs, not high performance designs.  The NC2030 is at the opposite end of the performance spectrum.

In any case, my objective is not an SDR, but a frequency selective wideband recorder and display which shows a waterfall of current activity in real time or a sped  up replay of recorded activity.  The recording would have to be fed to an SDR to demodulate it.  There is considerable advantage to storing each band separately when playing it back though an SDR.  It's about 2 TB per day, but at current disk prices, buffering the last 2-3 days of band activity is not very expensive.

The amateur HF allocation is about 10% of the HF spectrum.  Collecting and then discarding the other 90% of the HF may not be the best use of resources.

The use of the STM32F4xx and LPC4370 is an attempt to knock the BoM down by 1/2 or more.  The $350 was based on using a single high speed ADC.

As for the other thread, a person is making silly statements such as you can't do DSP without using analytic signals and that you must use complex series.  What is the point of linking that?

[ogden]

I cannot find the PDF nor the website

How convenient

The NC2030 is at the opposite end of the performance spectrum.

Thank you, did not know about this project. Impressive specs. You kind of prove your point.
Yet I can't find NC2030 in the radio table: http://www.sherweng.com/table.html
Worth to mention that FlexRadio which is 16-bit ADC direct RF sampling transceiver have #1 spot here

As for the other thread, a person is making silly statements such as you can't do DSP without using analytic signals and that you must use complex series.   What is the point of linking that?

- To show which thread or message you are talking about. Maybe readers of this thread want to read or say something there as well

[rhb]

It's a pretty stupid, off topic pissing contest, but as you insist:

https://www.eevblog.com/forum/microcontrollers/getting-dsp-basics/msg1924066/#msg1924066

I'm hoping the thread is dead and stays that way. I got sucked into when GeoffreyF said you *had* to use quadrature  to do DSP.   I think the source of the conflict was a BS EE view of DSP vs a PhD geophysics view which is rather broader than what someone learns at the undergraduate level in EE.  If you read the thread, you'll see that he did not get my point that you need to be able analyze a DSP problem with pencil and paper from knowledge of the common transform pairs to really be good at DSP. One of the great pleasures of my career was a period of several years when I had a *very* sharp Stanford grad across the hall.  We often walked in each other's office and started graphically analyzing various DSP problems on the white board as a preliminary step before implementing the idea.  Usually one of us would recognize some detail the other missed, so it was a great time saver. "No, that won't work because....".

When most of the people along the hall have PhDs from the absolute top schools you do *not* make comments such as this:

However, there are good engineering reasons for quadrature.  I suggest you find out what they are.

unless you want to be the butt of office jokes.  You damn well better rattle of the pros and cons of using quadrature in your second sentence.  And be prepared to articulately justify your remarks.  It is highly likely in that setting that the other person can do a detailed analysis on the whiteboard in a couple of minutes and may well know some bit of trivia that proves you wrong.

I'm extremely annoyed that I can't find the other PDF because the opposite image rejection is incredible.  I *really* need to reorganize my file system.  It's this God awful assortment of stuff spanning many years and many computers.  I've started on the chore a number of times, but eventually get so confused by several TB of duplicates  made worse by the fact that I'm actually running several computers plus a couple of VMs.

edit:  found the hard copy :-) cbjohn.com/aa0zz/PPLLUsers/W6JL/W6JL.pdf  My memeory was incorrect, it's only 60 dB of opposite side band suppression.  Still very impressive performance.


I've got one MSEE thesis and a PhD dissertation IIRC on the topic of mixers which provide a detailed analysis of quadrature mixers.  The PhD dissertation includes a detailed comparison of the performance of a number of mixers.

I'd guess that the reason the NorCal2030 doesn't appear in the list is lack of a unit to test.  They only assembled 100 kits and it's a rather complex radio.

I found out that some versions of the STM32F4 have ADCs which are plagued by on chip noise.

I can think of two likely problem areas for quadrature mixer builds, the output level and timing of the two ports and the linearity of the ADCs.

The problem with the big slurp is the data rate.  There is a really nice 16 bit SDR build in a recent QEX.  It's an eval board for a 125 or 150 MSa/S from AD and a ZedBoard plus packaging and software.  But that's a $1200 build.

I'm not interested in contesting nor in collecting wall paper, but an RPi + STM32F4 Discovery board HD display of 3.6 MHz of the HF band allocations seems to me would be a very attractive shack decoration for a contest or DX enthusiast.

Alberto, I2PHD has a nice STM32F429 based SDR:

http://www.sdradio.eu/weaksignals/armradio/

My personal interest is more in investigating propagation phenomena.

[xaxaxa]

If you can find a decent and cheap adc, the backend stuff (FPGA, USB interfacing) is very cheap; I have a standard xc6slx9+usb3343 layout that is a base for many of my projects that costs less than $10 in BOM costs and occupies about 5x5 cm of PCB area. The device shows up as a USB CDC (serial) device and can transfer about 30MB/s half duplex.

[ogden]

edit:  found the hard copy :-) cbjohn.com/aa0zz/PPLLUsers/W6JL/W6JL.pdf  My memeory was incorrect, it's only 60 dB of opposite side band suppression.  Still very impressive performance.

Similar, high complexity to NorCal2030. Note loads of trimpots in baseband filters. I wonder what would be complexity of your project - like this but multiplied by 8, hours/days of tuning/tweaking filters and IQ balance to get target performance? Seeming complexity of "multiband Tayloe receiver" is why I tend to promote RF sampling instead.

I've got one MSEE thesis and a PhD dissertation IIRC on the topic of mixers

Respect! 

There is a really nice 16 bit SDR build in a recent QEX.  It's an eval board for a 125 or 150 MSa/S from AD and a ZedBoard plus packaging and software.  But that's a $1200 build.

Eval boards are not the best components to build cost-sensitive device aimed at hobbyists

[rhb]

I would be using digital baseband filters, so quite a bit simpler.  What I plan is really just a pan adaptor on steroids.  The big advantage to all the complexity of those two designs is the low power consumption for the performance achieved.  You could never get that with an SDR.

The eval boards are the low cost option if you're only building one.  And some of the Analog Devices boards are just amazing, especially the ones in the $1200-1800 range.

You're overlooking the added expense of windowing to just show the ham allocations. While not too expensive for the waterfall, if you want to record for later replay it gets quite complex doing the BP filtering and subsampling.  Recording to disk presents a serious IO bottleneck if you don't window and subsample.  Recording just the US 160 to 1.25 m allocations requires about 2 TB/day at 12 bits.

I was mistaken.  What I thought was another thesis or dissertation is actually Soer's MS thesis:

https://essay.utwente.nl/58276/1/scriptie_Soer.pdf

Soer shows that the Tayloe with a 25% duty cycle has the best balance between noise figure and conversion loss.  I thought that the thesis covered regular diode mixers, but I see that's not the case.  It's just switching mixers.  But I may well have an analysis of diode ring and other mixer types at a similar level somewhere else.

I'm a heavy literature user.  I've got a personal 5000+ volume tech library and several 4 drawer file cabinets of papers.  I generally would not start the implementation of anything at work until I had pulled and read a couple dozen or more papers.  Generally I get some recent papers and then get the major references for a couple of layers down.

Literature access has become a huge nuisance. I'm in a small rural town in central Arkansas.   The only way I can get IEEE stuff is to make a 160 mile round trip to a university library and they limit my guest access to 2 hours per day.  Though my experience on my two visits  so far is that by the time I have spent 2 hours searching IEEE Xplore I'm pretty tired. 

The next time I do it, I plan to have already done the searches so all I'm doing is pasting citations from a file.  The IEEE appears to monitor what you're doing.  I've been kicked off a couple of times and had to log back in to continue.  I've not yet tangled with Elsevier.  I suspect they will be even more aggressive about limiting downloading.

[ogden]

The big advantage to all the complexity of those two designs is the low power consumption for the performance achieved.  You could never get that with an SDR.

For example LTC2164 (16 bit @105 MSPS) consumes 0.2W. I do not see power consumption as argument here, especially when you plan TV display and HDD recorder running.

The eval boards are the low cost option if you're only building one.  And some of the Analog Devices boards are just amazing, especially the ones in the $1200-1800 range.

I do not consider $1200-1800 board as low cost or amazing deal for hobbyist disregarding fact that he needs only one device in result. Thou we can agree to disagree here.

You're overlooking the added expense of windowing to just show the ham allocations. While not too expensive for the waterfall, if you want to record for later replay it gets quite complex doing the BP filtering and subsampling.  Recording to disk presents a serious IO bottleneck if you don't window and subsample.  Recording just the US 160 to 1.25 m allocations requires about 2 TB/day at 12 bits.

Are you saying that digital downconversion (DDC) is complex task? Note that in very first post I said "downconvert all subbands digitally in the FPGA". For those unaware - digital IQ baseband streams after 8 digital downconverters would be similar to those coming out of ADC's of 8 Tayloe detector-based receivers, will have identical IO bandwidth. I never said that whole 30MHz HF bandwidth shall be recorded.

[xaxaxa]

For example LTC2164 (16 bit @105 MSPS) consumes 0.2W. I do not see power consumption as argument here, especially when you plan TV display and HDD recorder running.

Still way too expensive
Let me know if you ever see a >=14 bit ADC with close to that sample rate for less than $10