Sounds pretty impressive Jeroen. Well done..
Walter/K5WH -----Original Message----- From: Jeroen Vreeken <[email protected]> Sent: Monday, April 6, 2020 6:28 PM To: [email protected] Subject: Re: [Freetel-codec2] Introducing a new FM mode: 6000 Hi David, I got LPCNet running at several different rates. I am not yet using it on the air though as the controller in our local repeater is an ARM system with not enough power to handle LPCNet (it is an dual core cortex-A7). I also found that in its current form it isn't very good at coding audio received from an analog QSO... For example this is a recording of our local net from this week: http://dmlinking.net/audio/2020/04/pi4za.202004050900.wav (In dutch, but the difference between original and through the codec is big enough) If I run it through LPCNet (even unquantised) the voicing/pitch never seems to be right. It might be interesting to know why. I can think of a few things that could interfering: - There is quite a bit of noise in the recording - It has been processed with the speex preprocessor to get rid of noise, maybe it introduces difficult artifacts? - The source material is already sampled at 8kHz, so the higher frequencies are already gone. - Is the equalisation off? On the bright side: I reshuffled some bits and now have room for 4800bps of voice data instead of 4667bps. 73s, Jeroen On 04/01/2020 07:45 AM, David Rowe wrote: > Nice work Jeroen, great to see you experimenting with a new voice mode! > > The LPCNet README: > > https://github.com/drowe67/LPCNet > > has some examples of using command line options to configure different > bit rates in the "Fully quantised encoder/decoder programs" section. > > -/- > > I am also WFH and have a few projects running. Mark, Bill and I are > working on a new low MDS telemetry system for High Altitude Balloons > (HAB) and other applications: > > https://github.com/drowe67/codec2/pull/101 > > and I'm also playing around with some Neural Net projects, including a > LPCnet decoder for FreeDV 700C. > > Plus I've been on the air with VK3RV and Jose LU5DKI using > KiwiSDR-assisted HF communications between Australia and Argentina > (7.177MHz +/- QRM 1030 UTC each evening). > > With bad things happening in the world it's a good time to stay home > and work on Ham Radio projects. > > Cheers, > David > > On 1/4/20 7:08 am, Jeroen Vreeken wrote: >> Hi all, >> >> During the last few weeks I had some more time available than >> expected (working from home for obvious reasons) and finally got >> around to do an experiment I wanted to do for a while now: implement >> and test a new mode for use on VHF/UHF. >> I really liked the idea of mode 2400B, but it is relativly wastfull >> with bandwidth, and I also wanted to have a bit higher quality codec2 >> mode so it can actually compete with analog FM. >> So I tried something different: a 6000baud mode with raised cosine symbols. >> A zero is encoded as an inversion, ones as a continuous level. >> To prevent to much DC in the signal after each 9 bits a zero is inserted. >> >> Originally I wanted to use a higher bitrate LPCNet mode, but couldn't >> figure out how to set it to anything other than 1733bps yet. So for >> the time being I am using mode 3200, but can still swap it out later >> as there is 4666bps available. >> I used a frame length of 120ms, partially to reduce overhead, but >> also to make it easier to switch to another codec later, 120ms can >> handle all the current codec sizes: 20ms, 30ms, 40ms etc... >> >> In order to make each identification more standardised and to always >> have a data channel available each frame has data. >> Either 64bits in case of a voice frame, or 608 in case of a full data frame. >> >> Yesterday I did some on-air tests between my home and our local 70cm >> repeater (the discriminator output is connected to the controller and >> can be sampled directly). At home I used the packet input of my ft-817. >> Using 5W this resulted in a near perfect (might need some finetuning) >> signal over a distance of 11km. >> >> My current status can be found at >> https://github.com/JeroenVreeken/codec2 in the m6000 branch. Be aware >> that I have not extensively tested with real audio data yet, my tests >> were with artificial voice data to be able to test easily. >> Next step is to use the new lib for the actual repeater. It is >> currently configured to receive both analog FM and mode 2400B, but it >> should not be hard to switch 2400B out for 6000. Just need to check >> that the correct amount of samples is used everywhere. That should >> enable some real world tests. >> >> 73s, >> Jeroen >> >> >> p.s. some values from the README: >> | second | frame >> ----------------|---------|------ >> baud | 6000 | 720 >> inserted 0s | 600 | 72 >> sync bits | 133.33 | 16 >> --------------------------------- >> voice bits | 4666.67 | 560 >> used voice bits | 3200 | 384 >> codec2 frames | 50 | 6 >> extra data bits | 533.33 | 64 >> control bits | 50 | 6 >> reserved bits | 16.67 | 2 >> --------------------------------- >> data frame bits | 5066.67 | 608 >> control bits | 83.33 | 10 >> reserved bits | | 14 >> >> >> >> >> _______________________________________________ >> 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
