I use a slightly different method of generating noise in software,
using the function
A   =   SQR(-2 * LOG(RND)) * SIN(2. pi. RND)  which gives a true
Gaussian spread with RMS equal to 1

Doubt it makes any difference in practice, apart from the wideband
noise spectrum and spread of the error function as bandwidth
approaches Fs/2.


Andy
www.g4jnt.com


On 3 February 2012 14:08, Rik Strobbe <[email protected]> wrote:
> Hello Andy,
>
> I just started from SNR (in dB) = 10*LOG(Psignal/Pnoise)
> Or for SNR = 1 (0dB) ==> Psignal = Pnoise, thus RMSsignal = RMSnoise
> If the signal is a sine (most obvious) then RMS of the signal = 
> 0.707*Amplitude of the signal
> If the noise is white (random function) then the RMS of the noise = 
> 0.577*Amplitude of the noise
> So for 0dB SNR Amplitude of the noise = 1.225*(Amplitude of the signal)
> As I sample at 11025Hz the total noise bandwidth is 5512Hz, and WSPR (what I 
> took as a reference) takes noise over 2500Hz BW I compensated for that.
> A bit to my own surprise by the first tests WSPR showed the same SNR as I 
> intended to generate
> (+/-1dB).
> As I intended to compare modes (not to make exact SNR measurement) I did not 
> worry too much about how correct the produced SNR was. My first idea was to 
> use that value as a start and then "fine tune" it with whatever WSPR came up 
> with. But that seemed not nessecary.
>
> Opera gives a 3-4dB offset for the SNR. maybe a noise bandwidth issue ?
> 5512Hz vs 2500Hz = 3.4dB ...
>
>
> 73,
>
> Rik
>
>
> ________________________________________
> Van: [email protected] [[email protected]] 
> namens Andy Talbot [[email protected]]
> Verzonden: donderdag 2 februari 2012 19:58
> Aan: [email protected]
> Onderwerp: Re: LF: QRSS3 "challenge"
>
> I was going to make muttering noises about the S/N being dependent on
> the QRSS speed, then realised   ..... duh....
> it doesn't matter!
>
> When normalised, -28dB always corresponds to 6dB above the noise bandwidth.
> What effective bandwidth (ie noise bandwidth divided by FFT bin size)
> have you assumed?
> Hamming window is about two FFT bins, Blackman Harris three,
> rectangular window is unity.
>
> Andy
> www.g4jnt.com
>
>
> On 2 February 2012 17:17, Rik Strobbe <[email protected]> wrote:
>> Dear all,
>>
>>
>>
>> as mentioned some days ago I generated a whole series of Opera, WSPR
>> and QRSS3 audio with a known SNR.
>>
>> For Opera and WSPR it was easy to determine the lowest SNR for a proper
>> decode. But for QRSS3 it is much more difficult as it depends on the
>> operators "sharp eye". It must be somewhere in the range of -24 to -28dB SNR
>> (@ 2.5kHz BW).
>>
>> Today I did put all the QRSS3 screenshots in this range in a
>> website, where you can fill in what you see (decode) for each screenshot.
>>
>> I would like to invite all of you to give it a try and send me the results
>> (the more entries the better the statistics).
>>
>> After some time (Feb 10th) will put all results into some nice tables
>> and graphs and make them available.
>>
>> Of course no personal (individual) results will be published. But based on
>> the results every participant (who keeps his results) can see where he ends.
>>
>>
>>
>> The "QRSS3 Challenge" can be found at http://on7yd.strobbe.eu/QRSS/
>>
>>
>>
>> 73, Rik  ON7YD - OR7T