Hi Alfred, > The signals I am talking about are at or just below the noise level. > Once the signal is +3dB or better then NR can make it more pleasant to > copy because I find it less fatiguing, but it certainly would be > copyable anyway. That is what I have concluded from making some tests.
> I don't think the FLEX system is any better or worse in this regard, as > it is a fundamental limit to signal processing. For SSB I think you are right, but not for CW, AM and FM. > A way of approaching it is to match the detection method with the known > characteristics of the signal. For CW that means selecting the proper > bandwidth, and the PSDR does a really good job with it's narrow filters. ?????????????? The narrowest filter I know how to get with PSDR is with the 25 Hz button and with specifying a DSP buffer of 4096. Sampling with 96 kHz (Softrock) I find a nice filter response that corresponds well to the buffer length. 4096 samples span a time of 42 milliseconds so the bandwidth should be in the neighbourhood of 1/0.042 Hz = 25 Hz if no care were given to stopband attenuation. Looking at the actual filter response gives these numbers: Level Width (dB) (Hz) -3 45 -6 70 -10 83 -20 115 -30 135 The skirts are steep and the stopband attenuation good so a 3dB bandwidth of 45 Hz is what I expect from theory. This filter will do an excellent job on crowded HF bands but it will not improve the detection of really weak CW signals in a background of white noise for the simple reason that the complete signal path also involves the human brain that has a filtering that is similar or even a bit narrower (for a well trained weak signal operator.) It is possible to improve significantly. I was using Linrad and predecessors during the 1990s on 144 MHz EME. (Look up old QST for the ARRL EME competition and compare scores to antenna size....) In Linrad one can set a filter that actually matches the CW bandwidth. In the range 14 to 18 Hz typically. The filters would be something like this: Level Width (dB) (Hz) -3 16 -6 17 -10 18 -20 19 -30 19.5 The buffer is huge and the associated time delay is about 1 second. Such a narrow filter must be exactly centered on the carrier and therefore an AFC (automatic frequency control) is needed. In Linrad with optimum EME parameters that would cause another 5 to 10 seconds delay. The filter would then be properly centered on a signal when a QSB peak occasionally brings a couple decodeable characters. The exact tuning would be based on the precise frequency of the CW carrier 5 or 10 seconds before as well as after the QSB peak. To copy CW in the optimized bandwidth we need S/N=8dB in e.g. 15 Hz. In 45 Hz, with PSDR or some even wider filter the effective bandwidth might be 40 Hz (including the human ear) and the signal could not be copied because the random fluctuations of the white noise would be stronger than the signal. Listening to the output is quite different from listening to a conventional filter such as the PSDR "25 Hz" filter. One can not hear any difference in frequency at all. It sounds as just a single frequency but the amplitude varies a bit. Noise and the CW signal sound exactly the same. Signals differ from noise just by being stronger and by having a pattern that we can recognize as CW. There is one further step we can take. It is possible to extract the CW carrier and then split the signal into I and Q. Normally I would be the CW signal while Q would be noise only. By listening to I only we can get rid of 50% of the noise and get a significant improvement. The carrier would be extracted from a filter with a bandwidth of perhaps 2Hz so it would be 17 dB above the noise on QSB peaks with a small phase error. In real life, at least during the 1990s, the frequency stability of most stations was inadequate so signals suffer from chirp. That means that the signal will start in both I and Q, pass over a pure I signal and end in Iand Q but with the opposite phase relation compared to the start. Then One can use a stereo head phone and utilize the excellent phase detection system in the human brain. There is still a significant improvement over just listening to the normal signal equal in both ears. It is a linear process and it supplies more information to the brain so it really helps. > For digital modes, such as PSK31 and JT65, the signal structure is known > and the decoding method can get significant processing gain so that the > raw signal can be well under the noise. Yes, and preprocessing will not help at all against white noise. (A sophisticated AFC to use when the qso partner is really unstable might add a feature that is not included in todays decoders.) Unlike CW where the detection (decision about 0 or 1) is in the human brain the digital modes do their decision in the computer. The methods are designed to be theoretically optimum for a signal in a background of white noise. If the interference is something else. Somewhat modulated spurs or pulses or something else that has a pattern that a computer could "learn" to recognize it will be possible to add algorithms that improve digital modes. Perhaps VERY significantly:-) > I wonder if some of the confusion may be between the different types of > 'noise'. The wideband thermal type is what NR is intended to reduce, Agree. In my mind the standard NR is just a cosmetic tool. It will improve the quality of signals that we can copy even if they have their original (bad) quality. > while pulse type QRN would be handled by the NB. The later can be > handled quite well with the right approach. I have read your excellent > papers on that topic. Yes. Linrad can do marvellous things on a calibrated VHF system:-) On HF where the frequency response of antennas and of the interference source become important Linrad is not so good. Different pulse sources would generate different pulse shapes. In principle this can be managed, but today Linrad requires that antennas and pulse sources have much more bandwidth than the SDR hardware. 73 Leif / SM5BSZ _______________________________________________ FlexRadio Systems Mailing List FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archives: http://www.mail-archive.com/flexradio%40flex-radio.biz/ Knowledge Base: http://kc.flexradio.com/ Homepage: http://www.flexradio.com/
