Hi Glen, I'm inclined to take a 3 tier approach for a VHF modems:
(1) FSK that can run through mic/spkr ports so people can get started with just a sound card and a $50 HT. This would require a good FSK modem that is robust to the vagaries of this channel. (2) GMSK that can be sub-optimally implemented on FM radios with data ports, or (3) optimally implemented with SDR. With (3) we'd get a large low signal performance jump over current VHF systems but require new, but low cost, open hardware. However I like this path rather than being limited by current FM radios. Timing estimation at these symbols rates can be done efficiently. On that note I'm working on a simple timing estimator and getting good results at Eb/No = 6dB in gmsk.m. Need to think about phase/frequency recovery next. For a first pass I might use a preamble, for GMSK that turns into a carrier at Rs/4. Then I'll test over a real VHF channel and see if the predicted performance for coherent GMSK compared to FM is for real. Overall I'd estimate a coherent GMSK modem is easier to get running on the STM32F4 than fdmdv, and that was straight forward. There is a remarkably simple implementation in that 1981 paper that uses 1-bit fixed point: flip flops and logic gates. Cheers, David On 21/12/14 10:35, glen english wrote: > Hi David > > I think whether one can do better than FM in the real world 100% depends > on exactly how you compare apples for apples. This is common mistake. > (not really a mistake, but not right) . > For people comparing digital to analog for 30dB SNR, well digital will > win every time. > But if comparing for analog = 12dB SNR, analog will often win over > commerical deployed technologies. > > Timing estimators and multiple modems it what I use these days. > > I'll estimate the clock and then run between 4 and 8 modem passes with > different timing, and pick the one that gives the lowest error > correction workload. > > Easy when you've got a 70K logic cell FPGA, but still possible on > STM32-4, I think you need to come up with a clever way to run the demod > sufficiently to decide what timing estimation is a decent candidate, and > then run the other 90% of the modem... (rather than what I do which si > run say 8 complete copies of the modem in parallel) . > > I think all DSTAR is incoherent, simple slicing (comparator) decoding. > Quite awful and suboptimal. > *** > For a common FM radio, you have : > 12dB octave high pass below 300Hz (corner at 330 Hz). > flat to 500Hz > 6dB per octave (roughly ) to 3kHz. (IE @ 1kHz you are 6dB up) . > Then brick wall at 3kHz. > The DC response will kill you trying to get GMSK without extra edge > generation. > Hence the OFDM modem does very well. > > For an FM radio adaptor, I'd encourage you to pursue a water filling > technique. > Maybe 125Hz bins > 1st Bin start = 375Hz >> 625 Hz, 64QAM (18 bits) > 750-1500 Hz = 16QAM (28 bits) > 1625-3125 Hz = QPSK (24 bits) > maybe....72 bits x 125 = 8750 bps > Golly that's quite a bit ! > > Transmit a PN or chirp at frame start for channel equalisation. Does not > have to be done very often as it is a BASEBAND equaliser, not a RF > channel equialiser. > > The 64QAM and 16 QAM would enable a hierarchical approach for those > constellations. > How you would map the bits to the codec important bits, I have not > thought about much. > > just ideas..... > > > > > > On 20/12/2014 1:32 PM, David Rowe wrote: >> Hi Glen, >> >> I just played around with the timing (Toff variable) in gmsk.m. At the >> 1% BER, Eb/No = 6dB point the demod BER hardly changes with +/-10% >> timing error, and the BER doesn't double (1dB implementation loss) until >> I plug in a timimg error of -30%, +20%. >> >> So I'd say we have a good shot at getting an ideal-performance timing >> estimator running. That 1981 paper suggest looking at the difference >> between the Re and Im arms, or we could even just try sampling at 10 >> different timing offsets and look at some statistic like the eye >> opening. I'm guessing that no raised root cosine filter makes life >> easier than PSK. >> >> I wonder if phase estimation is hard for GMSK?. I'm not even sure what >> DStar does for that, have to read the spec. There is some implied phase >> information in that logic at the end of the coherent demod, IIRC the way >> phases evolve can be modelled as OQPSK following some sort of state >> machine. Only certain state changes are allowed. >> >> I think it might be worth pushing this simulation through to the point >> where it can be tested over real VHF channels using stored files. We >> could then try using a regular FM radio and an SDR and see what sort of >> gain we can get over over FM. >> >> Not sure what to do about a transmitter ..... not sure if we can rely on >> a FM modulator to generate GMSK accurately. Maybe a SSB radio using >> upconverted stored files of GMSK? At 1200 bit/s that would be 1200 Hz >> wide. However we couldn't reproduce FM That way using the same tx/rx. >> >> I like the name G-star! >> >> Cheers, >> >> David >> >> On 20/12/14 11:13, glen english wrote: >>> Hi David >>> 1- Any FM demodulated process. If it's post demod, there will be the >>> threshold. >>> that's right - no one cares about low CNR 'cause it is unusable. >>> 2 - and you got to do carrier recovery very quickly for burst transmissions. >>> 3- I think good speech quality if the system allows for it at the VHF is >>> just as important as HF. Probably estimated SNR would be enough to >>> decide to use the extra bits. >>> It's a matter of coming up with a good scheme for codec2 to provide >>> better quality with say 2 x the bits. >>> 4- With a tighter rolloff, finding those clock spectral lines is always >>> going to be more difficult than if there is , at the extreme, zero rolloff. >>> >>> For my G-star system which I am fiddling with for VHF, I have your >>> codec2 @ 1000 bps , FM discriminator detection and block product turbo >>> coding. You need to deal with the clicks and improve threshold >>> performance for it to all be worthwhile, and the FEC satisfactory does that. >>> >>> I have not spent the time yet to sort your codec to provide 1) >>> hierachical performance and b) unequal protection of certain bits (IE I >>> have not spent the time to understand which bits are most important- >>> except for the obvious which is the MSBs are always more important than >>> the LSBs ! >>> >>> cheers >>> >>> >>> On 20/12/2014 11:24 AM, David Rowe wrote: >>>> Thanks Glan, I was hoping some one with a solid comms background would >>>> take a look at those recent posts. Some questions: >>>> >>>> 1/ Re the post-FM scenario, do you mean using FM radios with through >>>> their "data" (discriminator) port? Thanks, I had wondered what those >>>> detectors look like a low CNRs. Clicks don't sound like much fun. >>>> >>>> That might be why the non-coherent demod I described first in the post >>>> is popular - it works whenever the current FM radios do. So no one >>>> notices the loss of low CNR modem performance. >>>> >>>> 2/ Thanks for the tip on low CNR clock recovery, not something I've had >>>> to look at yet with MSK, although I've achieved good results with PSK >>>> modems. (G)MSK is pretty close to OQPSK. I might try some clock >>>> recovery simulations at low CNRs and see how we go. That 1981 paper had >>>> near ideal results in their practical modem 30 years ago, so suspect it >>>> can be done with a non-post FM (ie SDR) radio. We might need >>>> differential detection for phase recovery, I haven't looked into that yet. >>>> >>>> 3/ Yep, the hierarchical scheme is what I'm currently looking at for HF, >>>> in particular pushing the speech quality lower for poor channels. >>>> Thinking the opposite for VHF. With plenty of bandwidth on VHF we can >>>> just send a sub-carrier at say 6dB lower (even with GMS) for >>>> supplementary info. It could have a simple checksum or we could >>>> estimate SNR. >>>> >>>> 4/ What technical problem does MSK/RRC MSK make easier than GSMK? >>>> Agreed we've got plenty of bandwidth. >>>> >>>> 5/ I'm thinking of a couple of modems. One that gets through standard >>>> FM or data port radios, and a high perf SDR based mode. They could have >>>> the same bit rate and protocol, and the rx could auto-detect the >>>> waveform in use. >>>> >>>> Cheers, >>>> >>>> David >>>> >>>> On 20/12/14 10:04, glen english wrote: >>>>> Hi David >>>>> >>>>> Good analysis. I think though there needs to be some clarifications >>>>> (having done a few of these in my professional capacity) >>>>> >>>>> 1) 10dB over analog FM .?Not so fast ! I reckon 3dB worse at best. >>>>> You'll need to clarify whether you are using a a post FM detected >>>>> output, because the threshold effect will knock you over. >>>>> >>>>> 12dB SINAD wont occur much before 6dB CNR, depending on the limiter and >>>>> demodulator characteristics.. The clicks will be numerous and this will >>>>> knock out many bits, especially as you narrow the post detection >>>>> bandwidth, the click length will cover more symbols. >>>>> So the best you might do is on par with analog FM isf ising post >>>>> detected output. >>>>> >>>>> My guess is it will be 3dB worse than 25kHz FM for the first 6dB, and >>>>> then of course much much better. >>>>> >>>>> Even if you do demod pre the FM demod and limiter, clock recovery of >>>>> GMSK for coherent detection, especially down in the noise is HARD. In my >>>>> experience, this is the undoing. >>>>> >>>>> So GMSK is generally accepted as simple with a tradeoff. No free lunch. >>>>> Most engnieers come to the same conclusion. YOu limit and FM demod, >>>>> which is a simple receiver, you give away the low signal low CNR >>>>> operation. >>>>> >>>>> HOWEVER ! >>>>> We've got plenty of bandwidth,so why not just MSK, or RRC MSK ? There >>>>> is no need to close the eye with the tight filter, we've got no need for >>>>> narrow bandwidth >>>>> >>>>> That is unless you want to get it through a standard FM radio. >>>>> >>>>> I've long been a proponent (and in my private communicationss with you) >>>>> I've always been in favour of a hierarchical codec and modulation. >>>>> >>>>> That is the lower bit rate codec is encoded and transmitted very robust, >>>>> so when the goign is really tough, it still works. >>>>> Bits that provide the codec more precision are transmitted in a layer >>>>> that requires a higher SNR, and are used when available. I expect that >>>>> because of the no FEC nature, there will need to be some sanity check >>>>> on whether those higher order bits are useful. >>>>> >>>>> IE when the going is really tough, the quality is low but it works. >>>>> But then the signals are good, the quality improves. >>>>> This is the current advanatage of wideband FM (25kHz) over digital >>>>> systems. For most digital systems, we get stuck with the LCD. >>>>> >>>>> But you are correct, the GMSK does not easily lend itself to >>>>> hierarchical schemes. >>>>> >>>>> >>>>> glen >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> On 20/12/2014 7:22 AM, David Rowe wrote: >>>>>> This week I've been working on GMSK modem simulations: >>>>>> >>>>>> http://www.rowetel.com/blog/?p=3824 >>>>>> >>>>>> - David >>>>>> >>>>>> ------------------------------------------------------------------------------ >>>>>> Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server >>>>>> from Actuate! Instantly Supercharge Your Business Reports and Dashboards >>>>>> with Interactivity, Sharing, Native Excel Exports, App Integration & more >>>>>> Get technology previously reserved for billion-dollar corporations, FREE >>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=164703151&iu=/4140/ostg.clktrk >>>>>> _______________________________________________ >>>>>> Freetel-codec2 mailing list >>>>>> Freetel-codec2@lists.sourceforge.net >>>>>> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >>>>>> >>>> ------------------------------------------------------------------------------ >>>> Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server >>>> from Actuate! Instantly Supercharge Your Business Reports and Dashboards >>>> with Interactivity, Sharing, Native Excel Exports, App Integration & more >>>> Get technology previously reserved for billion-dollar corporations, FREE >>>> http://pubads.g.doubleclick.net/gampad/clk?id=164703151&iu=/4140/ostg.clktrk >>>> _______________________________________________ >>>> Freetel-codec2 mailing list >>>> Freetel-codec2@lists.sourceforge.net >>>> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >>>> >> ------------------------------------------------------------------------------ >> Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server >> from Actuate! Instantly Supercharge Your Business Reports and Dashboards >> with Interactivity, Sharing, Native Excel Exports, App Integration & more >> Get technology previously reserved for billion-dollar corporations, FREE >> http://pubads.g.doubleclick.net/gampad/clk?id=164703151&iu=/4140/ostg.clktrk >> _______________________________________________ >> Freetel-codec2 mailing list >> Freetel-codec2@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >> > ------------------------------------------------------------------------------ Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server from Actuate! Instantly Supercharge Your Business Reports and Dashboards with Interactivity, Sharing, Native Excel Exports, App Integration & more Get technology previously reserved for billion-dollar corporations, FREE http://pubads.g.doubleclick.net/gampad/clk?id=164703151&iu=/4140/ostg.clktrk _______________________________________________ Freetel-codec2 mailing list Freetel-codec2@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/freetel-codec2
