Re: [wsjt-devel] GFSK for FT8?
On 08/05/2019 09:18, Игорь Ч via wsjt-devel wrote: We should also take in account overlapping spectra as spreading information between symbols under GFSK will affect decoding of the overlapped signals mostly if we are using frequency criterion for candidate ranking and probably due to the inaccurate signal subtraction, assuming all signals on the band are GFSK modulated. Hopefully we can evaluate GFSK benefits/losses via simulation of the signals for overloaded band scenario. Hi Igor, Steve and Joe have done much of this work while designing the parameters for FT4. We have a considerable history of real World FT8 signals using MFSK modulation on which the GFSK decoder has been tested. We also have a multi-signal simulator which was used when characterizing the GFSK decoders. This is how the amount of smoothing applied to make GFSK signals was chosen, i.e. an amount of smoothing that reduced the number of successful decodes missed from a data set to an acceptably low level. There is no inaccurate signal subtraction when using the GFSK decoder to to subtract GFSK signals, the facsimiles generated for subtraction are exactly the same as the originally transmitted signals. The only inaccuracies are in the phase and amplitude and those errors should be unchanged from that measured by the old MFSK decoder. Perhaps you are saying that the accuracy of the phase measured from on air GFSK signals is lower than that measured from otherwise equivalent MFSK signals, if so I doubt that is significant. 73 Bill G4WJS. ___ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
Re: [wsjt-devel] GFSK for FT8?
Hi Bill, . We should also take in account overlapping spectra as spreading information between symbols under GFSK will affect decoding of the overlapped signals mostly if we are using frequency criterion for candidate ranking and probably due to the inaccurate signal subtraction, assuming all signals on the band are GFSK modulated. Hopefully we can evaluate GFSK benefits/losses via simulation of the signals for overloaded band scenario. . 73, Igor UA3DJY >Date: Sun, 28 Apr 2019 16:10:16 +0100 >From: Bill Somerville < g4...@classdesign.com > >To: wsjt-devel@lists.sourceforge.net >Subject: Re: [wsjt-devel] GFSK for FT8? >Message-ID: < 0dd82369-a042-b1fd-0ae0-c5acc91eb...@classdesign.com > >Content-Type: text/plain; charset="utf-8"; Format="flowed" >... >Lastly, going back to your initial question. I was probably a bit over >zealous in saying there are only compatibility issues and no sensitivity >issues. For a single signal communication in the presence of noise >alone, using GFSK will reduce sensitivity compared with FSK due to >information being spread between symbols (inter-symbol interference or >ISI for short). The aim is to use an amount of frequency change >smoothing so that, on a busy channel, the overall interference between >signals is reduced, by limiting their bandwidths, such that net >sensitivity is increased. Choosing the exact amount of smoothing to >apply is a complex problem since it depends on the number of signals, >the distribution of signal strengths, the frequency distribution of >signals, as well as the distribution of time-synchronization accuracies. >All of these characteristics vary continuously during real world FT8/4 >communication on the Amateur bands, so all we can really do is carry out >empirical tests and try to pick an optimum filter bandwidth which gives >best results for the whole community. > >73 >Bill >G4WJS. > ___ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
Re: [wsjt-devel] GFSK for FT8?
On 4/27/19 3:37 PM, Bill Somerville wrote: The approach we propose is to restrict the amount of smoothing applied to make a GFSK modulated signal to a relatively low amount and to perhaps apply an even smaller amount of smoothing to facsimile signals used for subtraction, so as to minimize the overall loss of sensitivity in a mixed FSK/GFSK environment. Obviously the intent is for all stations to move to GFSK modulation eventually as it has clear benefits despite the extra complexity of implementation, so any interim measures to limit sensitivity loss when decoding FSK modulated signals will be biased towards best performance for GFSK decoding. If we take that approach I assume the intention will be to move towards decoding purely optimized for GFSK modulation over time as users migrate. This in itself is not a compatibility issue, rather a quality of implementation issue. Fortunately the Gaussian function convoluted with a raw FSK signal to make GFSK is continuously variable so we can pick one that exactly suits our needs and vary it over releases as needed. The final choice of Gaussian smoothig functions for transmission and for signal subtraction has been made by doing empirical tests using multiple simulated FT8 signals and using sample data gathered on air. For example measuring how many potential decodes are lost decoding real-World FSK modulated signals using a decoder optimized for GFSK modulated signals. Fortunately for the levels of smoothing we propose this loss of sensitivity is relatively small as other effects like noise mixed with signals tend to dominate. Hi Bill and all, Some years ago, I have designed some RTTY TU's using tubes, transistors, IC's and OP amps. I have observed that the place where the optimization process has given the most remarkable improvement was often the post detection low-pass filter. On many popular designs, this filter has simply been neglected. Starting with the rather classical but conservative Bessel and gaussian based polynomials, I have observed that the limits of the possible improvement was quickly reached. But I have found that it's often better to tolerate a little bit of overshot, rather than to demand zero overshot. This tolerance of overshot has nearly no drawback but it offer the ability to obtain a more steep raising and falling edge which is an advantage for the following stage which produce finally a digital signal. This result is simply obtained by using a different polynomial, which is a little bit heuristic. Such a post-detection low-pass filter is different from a pre-modulation filter, about which one we are speaking here. But some similarities remain. I can imagine that a polynomial of the same type could help to improve the performances in a similar manner as it does in the case of a RTTY TU. Just an idea ! Best wishes, Claude (DJ0OT) ___ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
Re: [wsjt-devel] GFSK for FT8?
On 27/04/2019 18:06, Игорь Ч via wsjt-devel wrote: There are also implications for ramp up and down of the transmission boundaries since the GFSK filtering will naturally do this without extra implementation complications. Hi Igor, I was wrong in respect of the above. Steve, K9AN, pointed out that we actually do waveform envelope shaping, to limit the bandwidth of the transmission start and end, separately after converting to a GFSK signal. This is as a consequence of the way that we convolve the Gaussian function to smooth the frequency transitions which is all done in the frequency domain with padding symbols (later discarded in the FT8 case) added to the begin and end. The smoothing is done by convolving the instantaneous frequencies with a Gaussian across a three symbol window. The resulting frequency time series is converted to a phase angle time series, which in turn has the ramp up and down applied. The resulting signal is later used to generate the transmitted PCM audio stream. The envelope shaping is done by applying a raised cosine ramp up/down amplitude function to the phase angle time series signal at the beginning/end respectively. The FT4 waveform explicitly includes one symbol extensions at the beginning and end of the waveform with values equal to the first and last symbols respectively (the padding mentioned above). These two extra symbols have the ramp up and down applied across their full duration. For FT8 we need to maintain exact compatibility with the existing FSK waveform, at least in length and timing, so a different envelope shaping scheme is applied where the first 1/8 of the first symbol and the last 1/8 of the last symbol have a raised cosine ramp up/down function applied in the time domain. Lastly, going back to your initial question. I was probably a bit over zealous in saying there are only compatibility issues and no sensitivity issues. For a single signal communication in the presence of noise alone, using GFSK will reduce sensitivity compared with FSK due to information being spread between symbols (inter-symbol interference or ISI for short). The aim is to use an amount of frequency change smoothing so that, on a busy channel, the overall interference between signals is reduced, by limiting their bandwidths, such that net sensitivity is increased. Choosing the exact amount of smoothing to apply is a complex problem since it depends on the number of signals, the distribution of signal strengths, the frequency distribution of signals, as well as the distribution of time-synchronization accuracies. All of these characteristics vary continuously during real world FT8/4 communication on the Amateur bands, so all we can really do is carry out empirical tests and try to pick an optimum filter bandwidth which gives best results for the whole community. 73 Bill G4WJS. ___ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
Re: [wsjt-devel] GFSK for FT8?
Bill -- empirical tests using multiple simulated FT8 signals > Do you have software for us civilians that can do that? Create .wav files with multiple FT8 signals with controllable frequencies and strengths, I mean. SimJT doesn't, according to documentation https://physics.princeton.edu/pulsar/k1jt/SimJT_User.pdf. Thanks 73, Paul K6PO ___ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
Re: [wsjt-devel] GFSK for FT8?
Hi Bill, . Many thanks for the detailed answer. Absolutely agree with you on the need to increase FT8 band capacity, we will try to implement FT8 GFSK support too when it is done in WSJT-X. . 73, Igor UA3DJY Re: [wsjt-devel] GFSK for FT8? From: Bill Somerville - 2019-04-27 13:37:57 On 27/04/2019 13:01, Игорь Ч via wsjt-devel wrote: > Hello Joe, > . > Is there any possible losses, e.g. sensitivity or compatibility, at > pushing FT8 in software towards GFSK modulation? > . > 73, > Igor UA3DJY Hi Igor, the simple answer is yes. But only related to compatibility, for example if we elect to use more smoothing of frequency shift discontinuities then a decoder designed to be optimal for that will loose sensitivity when decoding old style FSK signals. We have some choices if we switch to GFSK for FT8, some more practical than others. We can go all out for best GFSK performance by matching the decoder to the new GFSK shaped signal and accept a loss of sensitivity when decoding old style FSK signals, we can try and detect the signal shape and adjust dynamically, and so on. One important area is multi-pass multi-signal decoding with subtraction of previously decoded signals, subtraction is most effective if the synthesized signal facsimiles to be subtracted are generated in exactly the same way that the original transmitted signal was. I.e. it is not optimal to generate a GFSK facsimile of a decoded message for the purpose of subtraction if the original was actually an FSK signal. The approach we propose is to restrict the amount of smoothing applied to make a GFSK modulated signal to a relatively low amount and to perhaps apply an even smaller amount of smoothing to facsimile signals used for subtraction, so as to minimize the overall loss of sensitivity in a mixed FSK/GFSK environment. Obviously the intent is for all stations to move to GFSK modulation eventually as it has clear benefits despite the extra complexity of implementation, so any interim measures to limit sensitivity loss when decoding FSK modulated signals will be biased towards best performance for GFSK decoding. If we take that approach I assume the intention will be to move towards decoding purely optimized for GFSK modulation over time as users migrate. This in itself is not a compatibility issue, rather a quality of implementation issue. Fortunately the Gaussian function convoluted with a raw FSK signal to make GFSK is continuously variable so we can pick one that exactly suits our needs and vary it over releases as needed. The final choice of Gaussian smoothig functions for transmission and for signal subtraction has been made by doing empirical tests using multiple simulated FT8 signals and using sample data gathered on air. For example measuring how many potential decodes are lost decoding real-World FSK modulated signals using a decoder optimized for GFSK modulated signals. Fortunately for the levels of smoothing we propose this loss of sensitivity is relatively small as other effects like noise mixed with signals tend to dominate. For FT4 none of this applies as we are using GFSK modulation from the get go and it is simply a case of selecting the amount of smoothing that achieves the wanted signal bandwidth characteristics. Note also that using GFSK modulation has an impact on transmission when changing a message on the fly which is considerably more complex as the whole waveform must be pre-calculated whereas with FSK modulation only the input symbols need be changed as the synthesized signal is generated instantaneously on the fly. There are also implications for ramp up and down of the transmission boundaries since the GFSK filtering will naturally do this without extra implementation complications. What is important is that moving to GFSK modulation *will increase overall channel capacity* so we hope any temporary loss of sensitivity will be outweighed by more potential concurrent decodes per bandwidth slot. This should be an attribute that is welcomed by HF users when occupancy is often approaching capacity at peak traffic times and at other times there will be no loss of performance once GFSK modulation is 100% adopted. 73 Bill G4WJS. >Суббота, 27 апреля 2019, 15:01 +03:00 от Игорь Ч : > >Hello Joe, >. >Is there any possible losses, e.g. sensitivity or compatibility, at pushing >FT8 in software towards GFSK modulation? >. >73, >Igor UA3DJY > -- Игорь Ч ___ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
Re: [wsjt-devel] GFSK for FT8?
On 27/04/2019 13:01, Игорь Ч via wsjt-devel wrote: Hello Joe, . Is there any possible losses, e.g. sensitivity or compatibility, at pushing FT8 in software towards GFSK modulation? . 73, Igor UA3DJY Hi Igor, the simple answer is yes. But only related to compatibility, for example if we elect to use more smoothing of frequency shift discontinuities then a decoder designed to be optimal for that will loose sensitivity when decoding old style FSK signals. We have some choices if we switch to GFSK for FT8, some more practical than others. We can go all out for best GFSK performance by matching the decoder to the new GFSK shaped signal and accept a loss of sensitivity when decoding old style FSK signals, we can try and detect the signal shape and adjust dynamically, and so on. One important area is multi-pass multi-signal decoding with subtraction of previously decoded signals, subtraction is most effective if the synthesized signal facsimiles to be subtracted are generated in exactly the same way that the original transmitted signal was. I.e. it is not optimal to generate a GFSK facsimile of a decoded message for the purpose of subtraction if the original was actually an FSK signal. The approach we propose is to restrict the amount of smoothing applied to make a GFSK modulated signal to a relatively low amount and to perhaps apply an even smaller amount of smoothing to facsimile signals used for subtraction, so as to minimize the overall loss of sensitivity in a mixed FSK/GFSK environment. Obviously the intent is for all stations to move to GFSK modulation eventually as it has clear benefits despite the extra complexity of implementation, so any interim measures to limit sensitivity loss when decoding FSK modulated signals will be biased towards best performance for GFSK decoding. If we take that approach I assume the intention will be to move towards decoding purely optimized for GFSK modulation over time as users migrate. This in itself is not a compatibility issue, rather a quality of implementation issue. Fortunately the Gaussian function convoluted with a raw FSK signal to make GFSK is continuously variable so we can pick one that exactly suits our needs and vary it over releases as needed. The final choice of Gaussian smoothig functions for transmission and for signal subtraction has been made by doing empirical tests using multiple simulated FT8 signals and using sample data gathered on air. For example measuring how many potential decodes are lost decoding real-World FSK modulated signals using a decoder optimized for GFSK modulated signals. Fortunately for the levels of smoothing we propose this loss of sensitivity is relatively small as other effects like noise mixed with signals tend to dominate. For FT4 none of this applies as we are using GFSK modulation from the get go and it is simply a case of selecting the amount of smoothing that achieves the wanted signal bandwidth characteristics. Note also that using GFSK modulation has an impact on transmission when changing a message on the fly which is considerably more complex as the whole waveform must be pre-calculated whereas with FSK modulation only the input symbols need be changed as the synthesized signal is generated instantaneously on the fly. There are also implications for ramp up and down of the transmission boundaries since the GFSK filtering will naturally do this without extra implementation complications. What is important is that moving to GFSK modulation *will increase overall channel capacity* so we hope any temporary loss of sensitivity will be outweighed by more potential concurrent decodes per bandwidth slot. This should be an attribute that is welcomed by HF users when occupancy is often approaching capacity at peak traffic times and at other times there will be no loss of performance once GFSK modulation is 100% adopted. 73 Bill G4WJS. ___ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
[wsjt-devel] GFSK for FT8?
Hello Joe, . Is there any possible losses, e.g. sensitivity or compatibility, at pushing FT8 in software towards GFSK modulation? . 73, Igor UA3DJY ___ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel