Topband: FT8 - How it really works

[Joe Subich, W4TV]

> Is the definition of "noise floor" being changed for FT8?
WSJT-X (and WSJT before that) defines noise as the integrated value
of noise (noise power) across the 2500 Hz (approximately based on
the receiver filter) receive bandwidth.

73,

    ... Joe, W4TV


On 2018-12-19 9:57 PM, K4SAV wrote:
> Joe, thanks for the information.  I am not exactly sure what all that 
> means. My conclusions were based on observed data.  It seems pretty 
> obvious to me that a signal that is more than 50 dB above the noise 
> floor should not receive a S/N number of -1 dB, which is what FT8 
> gives.  I don't know how the information you provided can make a 
> calculation like that.
> 
> I judge that a signal reading S9+40 dB on the S meter should be more 
> than 50 dB above the noise floor when I can tune of to a spot where 
> there are no signals and the S meter reads about S2 or S3 in SSB mode or 
> less than S1 in a narrow bandwidth.  Is the definition of "noise floor" 
> being changed for FT8?
> 
> Jerry, K4SAV
> 
> On 12/19/2018 7:27 PM, Joe Subich, W4TV wrote:
>> On 2018-12-19 4:28 PM, K4SAV wrote:
>> > The official documentation for FT8 says it will decode signals 24 dB
>> > below the noise floor.  That is not a correct statement most of the
>> > time.
>>
>> No, that is a correct statement.  Signal reports in WSJT-X for FT8, JT65
>> and JT9 are *all* measured *with regard to the noise in 2500 Hz*. Note
>> that the tone filters in WSJT-X are on the order of less than 12 Hz or 
>> so wide so the SNR *for an individual tone in the DSP filter bandwidth*
>> at 0 dB is -23 dB relative to the *total noise in 2500 Hz bandwidth*.
>> The actual filter bandwidth will change from mode to mode due to the
>> differences in keying rated and tone spacing ... the actual SNR limit
>> is shown in section 17.2.7 of the WSJT_X 2.0 User Guide.
>>
>> CW operators understand this from experience ... a quality 200 Hz filter
>> will have ~12 dB less noise than a 2800 Hz filter.  Thus a CW signal
>> with a 200 Hz filter will have 12 dB better SNR than the same CW signal
>> with a 2800 Hz filter (excluding any "processing gain" from the ear-
>> brain filter).
>>
>> With FT8, JT65, JT9, etc. coding (forward error correction) provides
>> some additional SNR (called "coding gain") but the *measurement* is
>> based on strength of the individual tone to total noise.  Thus, the
>> lowest accurate report is -24 dB although some signals will be decoded
>> at levels below that.
>>
>> 73,
>>
>>    ... Joe, W4TV
>>
>>
>> On 2018-12-19 4:28 PM, K4SAV wrote:
>>> While sitting around being bored and recovering from a gall bladder 
>>> operation, I decided to do some experiments with FT8.  First thing I 
>>> did was upgrade the software to WSJT-X v2.0.
>>>
>>> I hope this post doesn't turn into another FT8 bashing session. My 
>>> only goal was to understand how this mode works and what it can do 
>>> and what it cannot do.
>>>
>>> The official description of FT8's signal reporting cannot be correct. 
>>> It is obviously not a signal to noise number and it is not an S meter 
>>> reading.  What is it? That was the first question to answer.  It's 
>>> obviously not an S/N number because how do you give a report of -1 dB 
>>> for a signal that is S9+40 dB on a quiet band.  I was unable to find 
>>> any info on how the signal report was calculated so I tried to 
>>> correlate those reports to observations.
>>>
>>> I think I have figured out a method that results in very close to the 
>>> same number that FT8 reports.  Here is the experiment.  I set up my 
>>> main VFO to USB 2500 Hz bandwidth and set the second VFO to CW at 
>>> about 150 Hz bandwidth.  I look for a station calling CQ and tune the 
>>> second VFO to him and measure his signal strength.  I also look at 
>>> the S meter for the signal level on the main VFO.  I also look at the 
>>> signal report calculated by the software.  For stations calling CQ 
>>> that report is calculated by the software in my computer.
>>>
>>> The FT8 report is usually very close to the difference in signal 
>>> levels (VFO1 - VFO2).   For example if the main VFO reads S9+10 and 
>>> the second VFO reads S9, the FT8 number will be -10 dB. Note that the 
>>> FT8 says that -24 dB is the lowest it can decode. With VFO1 = S9+10, 
>>> that's about S7 for the smallest signal it can decode.  Observations 
>>> agree. Those numbers will vary a little depending on how your S meter 
>>> is calibrated.  In order to decode a weak signal, all those close USA 
>>> stations will have to go silent.
>>>
>>> The official documentation for FT8 says it will decode signals 24 dB 
>>> below the noise floor.  That is not a correct statement most of the 
>>> time.  That statement should be that FT8 will decode signals 24 dB 
>>> below the sum total of everything in a 2500 Hz bandwidth. If the 
>>> total of all signals on the band are below the noise floor, it would 
>>> be interesting to know if FT8 will decode any of them.  I haven't 
>>> observed that yet in a real situation. I did however try to simulate 
>>> that condition by adding enough noise to the signals such that all 
>>> the signals were below the noise.  The software did continue to 
>>> decode signals.  All the reports were -24 dB.  This was a very crude 
>>> test because I don't know how exactly much the signals were below the 
>>> noise.  This should be of benefit to those people that have S9+ noise 
>>> on the bands they operate. They should be able to decode the 
>>> strongest signals on the band.
>>>
>>> The (VFO1 - VFO2) test just described should always result in a 
>>> number equal to or less than zero.  I notice sometimes the software 
>>> will report a small positive number.  That seems to happen more often 
>>> when the bandwidth is set to something less than 2500 Hz and there 
>>> are very few signals on the band.  I think this may be related to the 
>>> fact that FT8 does all its calculations using audio signals and the 
>>> receiver S meter is operating on RF. Audio shaping in the receiver 
>>> will affect the FT8 calculations. Audio processing in your computer 
>>> sound card may be a factor too. This becomes really apparent when the 
>>> radio is set to CW and the audio peaking filter is turned on.  With 
>>> SSB bandwidth and flat audio response, S meter readings are a good 
>>> indication of what will be decoded.  It should decode signals down to 
>>> 24 dB below whatever your S meter reads.
>>>
>>> I also narrowed the bandwidth of VFO1 and chopped out a bunch of 
>>> signals.  I got S7 on VFO1.  Then a station calling CQ also measured 
>>> S7 on VFO2.  The FT8 report was 0 dB.  Agrees.`
>>>
>>> That test brings up a possibility.  If you can narrow VFO1 to a very 
>>> narrow bandwidth hopefully containing only a very weak signal, then 
>>> you may be able to decode it.  A strong signal in the passband of 
>>> VFO1 will kill the decode.
>>>
>>> It works.  I decreased the bandwidth of VFO1 to 200 Hz and it decoded 
>>> an S2 signal.  I had VFO1 in USB mode with that bandwidth. My 
>>> receiver will go to zero bandwidth in USB mode.  I put VFO1 into CW 
>>> mode at 100 Hz bandwidth and it decoded a signal that was moving the 
>>> meter between S0 and S1.  That signal would have also been easy copy 
>>> if it was CW instead of FT8.  I was using a good receiving antenna on 
>>> 160 meters immediately after sunset.
>>>
>>> While this seems to work for weak signals it is a non-starter for 
>>> normal operation.  How do you tune around with a very narrow 
>>> bandwidth looking for a station calling CQ or any other station that 
>>> might be DX?  It's not like CW, unless you learn to copy FT8 by ear. 
>>> You can't find him with a wide bandwidth because the software won't 
>>> decode him.  He is only there when the bandwidth is very narrow. 
>>> Given the number of USA stations on FT8 that bandwidth will have to 
>>> be really narrow to keep the USA stations out of the passband.  Even 
>>> 50 to 100 Hz bandwidth usually doesn't do it on a crowded band and 
>>> you can't go lower than that and still decode the signal.  This 
>>> doesn't sound like anything that is practical.  Maybe something 
>>> useful might be to improve the copy of a weak station by narrowing 
>>> the bandwidth if you already know the station is there.
>>>
>>> One thing you could do is set the receiver to a narrow bandwidth and 
>>> call CQ DX, listening only on your transmit frequency. However the DX 
>>> station would probably need to be receiving with a very narrow 
>>> bandwidth or he won't hear you because you are probably very weak on 
>>> his end too. I seriously doubt that he knows to do that because it 
>>> seems that no one else knows about that either.  Besides it is not 
>>> often that a rare DX station will respond to a USA station calling CQ 
>>> DX.  Another non-practical suggestion.
>>>
>>> There are DX stations strong enough to be decoded that can be worked 
>>> with FT8, especially on the higher bands like 20 meters. Even on 160 
>>> meters sometimes a DX station will be strong enough to be decoded. 
>>> Just tonight right after sunset I heard a couple of European stations 
>>> on 160 running S5 to S6.  Because they were so strong, I tuned down 
>>> to the CW portion of the band but I didn't hear a single signal from 
>>> anyone down there.  Oh well.
>>>
>>> Seems to me that FT8 is a very poor method of working weak signal DX. 
>>> It also seems that it isn't being used that way either.  Just 
>>> listening, it seems that everyone is working very strong signals, 20 
>>> to 40 dB above the noise floor, at least as observed at my station. 
>>> Maybe this isn't the case for people that have an S9+ noise floor. 
>>> For those people, if they can't reduce the noise, FT8 may be the only 
>>> way they can do any operating.
>>>
>>> At least I now know more about FT8 than I did before starting this 
>>> exercise.  Learning stuff is never boring and it killed some time, 
>>> and my big incision feels a little better.
>>>
>>> Jerry, K4SAV
>>> _________________
>>> Searchable Archives: http://www.contesting.com/_topband - Topband 
>>> Reflector
>>>
>> _________________
>> Searchable Archives: http://www.contesting.com/_topband - Topband 
>> Reflector
> 
> _________________
> Searchable Archives: http://www.contesting.com/_topband - Topband Reflector
>