Hi Take,

Your textbook figure shows the difference in Eb/N0 required to achieve 10^-4 
BER on the AWGN channel, using single-symbol non-coherent detection. 

Please consider the fact that if the SNR is large enough to provide 10^-4 BER, 
then any of the JT modes would virtually never fail to decode. We would be 
operating like a commercial communications system, at SNR well above the 50% 
decoding threshold that is used to characterize the relative sensitivity of the 
different JT modes. 

Consider FT8 as an example - the error-correcting code that we use will decode 
about half of the frames having 26 bit errors (out of 174 total bits). This 
means that our 50% decoding threshold is obtained at a BER of approximately 
26/174=0.15. This is more than 3 orders of magnitude larger than the BER in 
your textbook Figure. 

At the BER corresponding to the FT8 detection threshold, you will find that the 
advantage of 64FSK over 8FSK is relatively small, as I stated earlier.  

As I also stated earlier, any near-threshold-SNR advantage of 64FSK can be 
completely erased if the 8FSK signal is detected on the basis of groups of 
symbols using a technique called “noncoherent sequence detection”.  The 
efficacy of sequence detection is not just something that I read about in a 
textbook.  This idea has been implemented in the improved WSPR decoder that was 
released in version 1.8, where the detection threshold has been improved by 
almost 3 dB for highly coherent MF and LF signals.

Finally, you have not commented on the comparison that I presented earlier 
where I scaled QRA64 to the same total transmission time as FT8, and I showed 
that their detection threshold were virtually the same. If 64FSK has a large 
advantage of 8FSK at the relevant SNRs, then how do you explain the parity 
between scaled QRA64 and FT8?  

My assertions are not based on textbook figures. They are based on results 
obtained from realistic simulations, using detection and decoding algorithms 
implemented in computer programs written by Nico, Joe, and me. You have 
asserted that it should be possible to send nearly 5 times as many bits as FT8 
sends, in the same amount of time, while improving the decoding threshold by up 
to 1 dB over the current FT8 threshold. As Joe has already said, we invite you 
to prepare some code that demonstrates your assertion. 

73, Steve k9an

> On Jul 5, 2018, at 10:39 PM, Tsutsumi Takehiko <ja5...@outlook.com> wrote:
> 
> Hi Steve,
>  
> This is a result of my searching “modulation theory book” in my shelves.
>  
> See the attached copy and Figure 1.1 in Japanese.
>  
> It indicates the degradation of EB/N0 by the increase of M concerning M-FSK 
> and this curve contradicts your “the energy-per-bit is the same, either way” 
> in your previous memo.
>  
>  
> Regards,<5CCC4EC6C6A94E25B2652B3E05EAC86F.jpg>
>  
> take
>  
> de JA5AEA
>  
> Sent from Mail <https://go.microsoft.com/fwlink/?LinkId=550986> for Windows 10
>  
> From: Tsutsumi Takehiko <ja5...@outlook.com <mailto:ja5...@outlook.com>>
> Sent: Wednesday, July 4, 2018 10:20:16 PM
> To: WSJT software development
> Cc: Steven Franke
> Subject: RE: [wsjt-devel] Observation on Expedition Mode
>  
> Steve,
>  
> Thank you for your comments.
>  
> Ok, I understand what is the disagreement between us.
>  
> First, I do not fully understand the point why you object my +4dB gain but I 
> need to find modulation theory book in my book sheves (never opened for more 
> than 20 years) and check about “ *bit* SNR, Eb/N0” related area before my 
> final response. If I can not solve for myself, I can call my local experts 
> without bothering you anymore.  But, I am thinking the longer symbol proposal 
> does not heart any RF performance including the delay spread, fading 
> performance, interference performance and a fraction of a dB gain with 
> non-coherent detection given from you is another goody.
>  
> Concerning the “second +4dB term”, I remember Bill’s suggestion included 
> hashing of callsign (I used the number of (28-15) x 5 = 65 bit reduction). I 
> also reduced five 7x7 synch to one 7x7 synch, which I am not sure whether 
> Bill included. I will not further describe the details to defend my ballpark 
> number but I can say I disagree with your 0.6dB gain calculation, here.
>  
> Thank you very much spending your precious time to comment on my +8dB 
> challenge.
>  
> Joe,
>  
> As I write to Steve, I do not have any intention to sell or stick to my +8dB 
> number. If you feel uncomfortable, please deal it to + (1~8)dB  additional 
> gain proposal .
>  
> take
>  
> de JA5AEA
>  
> Sent from Mail <https://go.microsoft.com/fwlink/?LinkId=550986> for Windows 10
>  
> From: Steven Franke via wsjt-devel <wsjt-devel@lists.sourceforge.net 
> <mailto:wsjt-devel@lists.sourceforge.net>>
> Sent: Wednesday, July 4, 2018 6:32:00 PM
> To: WSJT software development
> Cc: Steven Franke
> Subject: Re: [wsjt-devel] Observation on Expedition Mode
>  
> Hi Take-san,
> 
>> I meant QRAXX as “Q-ary Repeat-Accumulate Codes for Weak Signal 
>> Communications” in  Nico’s literature but I do not have any intent to modify 
>> wsjt-X “QRA64” mode to this discussion.
> 
> Understood. But why not scale the well-known results from Nico’s excellent 
> QRA64 mode to see what should be possible?
> 
>>  
>> By the increase of symbol length to 64mS (FT64) and 74.7mS (FT128) from 32mS 
>> (from 160mS FT8 symbol length speed up by factor 5), the gain is 
>> 10LOG(64/32) = +3dB and 10LOG(74.7/32) = +4dB. 
> 
> You have shown that the *symbol* SNR (Es/N0) will be doubled if we use 64FSK 
> instead of 8FSK, but what matters is the *bit* SNR, Eb/N0. 
> 
> A single 64FSK symbol conveys 6 bits of information, whereas the 8FSK symbol 
> conveys only 3 bits. You neglected to factor this into your calculations. 
> While the 64ms 64FSK symbol contains twice the energy of the 32ms 8FSK 
> symbol, the energy-per-bit is the same, either way. 
> 
> There *is* a slight modulation-detection-efficiency advantage to using 64FSK 
> instead of 8FSK if the symbol detection is done noncoherently on a 
> symbol-by-symbol basis, but the gain is fraction of a dB.  Furthermore, any 
> such advantage vanishes if the 8FSK demodulator is configured to detect 
> sequences of, say, 2 or 3 symbols rather than individual symbols.
> 
> In any case, the 64FSK vs 8FSK advantage was already included in the scaled 
> QRA64 example that I described earlier.  I stand by my conclusions. 
> 
> I also fully agree with Joe’s objection to your "second" +4dB term. Each FT8 
> message conveys 75 bits. If we send five of them serially, then we are 
> sending 375 bits.  If we were to combine the essential information into a 
> single packet, it would need to convey a total of 11 callsigns (the Fox call, 
> 5 calls to be printed with RR73, and 5 calls to be printed with signal 
> reports_ plus 5 signal reports. This is a total of 11*28 + 5*3 where I have 
> conservatively used only 3 bits for each signal report. Thus, the available 
> savings is 10*log10(375/323) = 0.6 dB. 
> 
> 73, Steve k9an

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Reply via email to