Nice work Jeroen! Your results (1% BER at 12dB Eb/No) are very similar to the ME-2FSK (a similar baseband PAM over analog FM scheme) in the table in this post:
http://www.rowetel.com/wordpress/?p=4663 I'm actually surprised this style of modem works so well, only about 3dB off an ideal 2FSK demod. You can the minimum detectable signal (MDS) in dBm for your waveform with: MDS = EbNodB + 10*log10(Rb) + NF - 174 where NF is the noise figure of your Rx. Cheers, David On 11/4/20 6:28 am, Jeroen Vreeken wrote: > So here are some first BER experiments. > I made a few small improvements to the demodulator (thanks to the > measurements). > But basicly it is still the same. > I used the fm.m modulator and demodulator. > > ber > > The amount of measurement points is still a bit limited, and since I > only use about 24seconds of audio data the numbers are not realy > accurate as of 16dB (from then on the BER is 0, since there are around > 1e+5 bits in the file that makes sense). > Since fm.m introduces noise compared to the carrier this is not a true > Eb/N0 value. > If I am correct the defaults for fm.m limit the bandwidth to 16kHz so > the Eb/N0 should be 4.2dB higher for a 6000 bps rate. > > A funny coincidence: shifting this plot 4.2 db more or less gives the > same plot as David got for his Bell 202 simulation. > Except ofcourse that this is 5 times more bits going through the same > channel. > > ber ebn0 > > I am going to do some more measurements: I want to see the effects of > scrambling and I also want to try some different pulse shapes. > The current shape is a raised cosine in the time domain, but a raised > cosine in the frequency domain should allow a smaller bandwidth. Since > the fm.m demodulator already cuts of most of the high frequencies it > seems we don't really need them and not sending them would allow more > energy to go into the usefull data. > Btw, the article 'The Shape of Bits to Come' > (https://www.qsl.net/n9zia/a108/index.html) was some great inspiration > to me years ago when I was trying to decode cubesat data. > > 73, > Jeroen > > > On 04/08/2020 12:00 AM, David Rowe wrote: >> This simulation: >> >> https://github.com/drowe67/codec2/blob/master/octave/fsk.m >> >> Was used to compare non-coherent 2FSK to a AFSK over FM, it includes an >> analog FM modem simulation fm.m (equivalent of a VCO/discrimator). The >> fm.m module has optional voice band filters. >> >> mancyfsk.m is similar, but with manchester encoder baseband signal, >> which I think is similar to what is being used in this work. >> >> Reason I mention this is that the best way to compare the effect of any >> signal processing step (like a scrambler) is to measure the bit error rate. >> >> Cheers, >> David >> >> On 8/4/20 7:12 am, Jeroen Vreeken wrote: >>> Indeed, I use the packet port of my ft-817, so the signal goes almost >>> directly to the oscilator and almost directly from the discriminator. >>> It might also work over the audio ports as the bandwidth might fit if >>> the filters are not to tight. >>> But then you would have to compensate for the pre/de-emphasis. I haven't >>> tried that yet. >>> >>> The bit stuffing is indeed to make sure we don't go near DC, and to make >>> sure there are enough 'flips' to recover the bit timing. >>> The scrambler looks nice, but we would have to find a balance between >>> lots of zeroes (easy to sync) and a perfect spectrum. >>> I'll try some variants on the air soon. >>> Also note that the frames generated with the test program are not >>> varying much, a stream with real-world data might look different. But >>> the pictures you showed both don't look that bad... >>> >>> >>> On 04/07/2020 11:24 PM, Steve wrote: >>>> I believe it is just direct FSK of the FM oscillator. +/- some >>>> deviation. Using the data port of most modern VHF radios. >>>> So on receive you just slice the ratio detector (discriminator output). >>>> >>>> The bit-stuffing keeps from long runs of a high or low. My experiment >>>> is to replace bit stuffing with a scrambler, but I haven't found all >>>> the places to remove the bit-stuffing :-) >>>> >>>> On Tue, Apr 7, 2020 at 4:08 PM David Rowe <[email protected] >>>> <mailto:[email protected]>> wrote: >>>> >>>> Hi Steve, >>>> >>>> >From what I understand (please feel free to correct me Jeroen) >>>> this is a >>>> baseband PAM waveform over analog FM. I'm haven't seen any >>>> theoretical >>>> analysis of these schemes, but we did measure some a few years back: >>>> >>>> http://www.rowetel.com/wordpress/?p=4663 >>>> >>>> Cheers, >>>> David >>>> >>>> On 7/4/20 5:58 pm, Steve wrote: >>>> > What would be expected here? Say about 9.5 dB Eb/N0 at 10E-5? I >>>> think >>>> > that's what the G3RUH modem got. >>>> > >>>> > It will definitely fill up the channel! :-) >>>> > >>>> > On Tue, Apr 7, 2020 at 2:43 AM David Rowe <[email protected] >>>> <mailto:[email protected]> >>>> > <mailto:[email protected] <mailto:[email protected]>>> wrote: >>>> > >>>> > Steve and Jeroen - it would be interesting to measure the >>>> modem BER (or >>>> > Packet Error Rate) performance. Steve - we could use >>>> cohpsk_ch to >>>> > inject measured amounts of noise. >>>> > >>>> > Cheers, >>>> > David >>>> >>>> >>>> >>>> >>>> >>>> _______________________________________________ >>>> Freetel-codec2 mailing list >>>> [email protected] >>>> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >>> >>> _______________________________________________ >>> Freetel-codec2 mailing list >>> [email protected] >>> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >>> >> _______________________________________________ >> Freetel-codec2 mailing list >> [email protected] >> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >> > > > > _______________________________________________ > Freetel-codec2 mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/freetel-codec2 > _______________________________________________ Freetel-codec2 mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/freetel-codec2
