Re: [Signal PKG] limited slope

[Renato S. Yamane]

2017-12-18 4:05 GMT+01:00 Ozzy Lash <address@hidden>:
> On Sun, Dec 17, 2017 at 1:13 PM, Renato S. Yamane wrote:
>> 2017-12-17 18:48 GMT+01:00 Robert T. Short
>> > On 12/16/2017 06:25 AM, Renato S. Yamane wrote:
>> >> 2017-12-15 16:34 GMT+01:00 Robert T. Short
>> >>> On 12/15/2017 12:27 AM, Renato S. Yamane wrote:
>> >>>> 2017-12-14 20:25 GMT+01:00 Robert T. Short
>> >>>>> On 12/13/2017 11:39 PM, Renato S. Yamane wrote:
>> >>>>>>
>> >>>>>> Please, see the frequency response on image available in:
>> >>>>>> https://ibb.co/cgwOqG
>> >>>>>>
>> >>>>>> Detail: when I use a crest factor of 12dB instead 6dB, I don´t have
>> >>>>>> this problem.
>> >>>>>>
>> >>>>>> ==============
>> >>>>>> pkg load signal;
>> >>>>>> pkg load ltfat;
>> >>>>>>
>> >>>>>> sampling_rate = 44100;
>> >>>>>> lenght = 30;
>> >>>>>> hpf = 400;
>> >>>>>> lpf = 4000;
>> >>>>>> crest_factor = 6;
>> >>>>>>
>> >>>>>> typenoise = noise((lenght)*sampling_rate, 1, 'pink');
>> >>>>>> [b,a] = butter(2, [hpf/(sampling_rate/2), lpf/(sampling_rate/2)]);
>> >>>>>> filtered = filter(b, a, typenoise);
>> >>>>>> filtered = filtered / (rms(filtered) / 10^(-crest_factor/20));
>> >>>>>> audiowrite ('signal.wav', filtered, 44100);
>> >>>>>> ==============
>> >>>>>
>> >>>>>
>> >>>>> I don't get the same spectral plots you show here.  My results are
>> >>>>> much
>> >>>>> more
>> >>>>> like your "expected" slope.
>> >>>>
>> >>>>
>> >>>> Wowww!
>> >>>> Can you tell me how you plot it? Just to be possible I make the same
>> >>>> check here in my side...
>> >>>
>> >>>
>> >>> Well, I used a Welch's method estimator to get the spectral density.
>> >>> I
>> >>> used
>> >>> a Hamming window, but just about any window should do.  I didn't use
>> >>> any
>> >>> overlap.  There is such an estimator as part of octave.  Then I just
>> >>> plotted
>> >>> it.  I did a semilogx just like you did and only for frequencies >
>> >>> 1Hz.
>> >>
>> >>
>> >> Sorry, but can you help me with this "coding" that you did?
>> >>
>> >>
>> > [Tn,f] = pwelch(typenoise, hamming(2^12));
>> > [Fn,f] = pwelch(filtered, hamming(2^12));
>> >
>> > idx = f>0;
>> > figure(1);
>> > semilogx(f(idx),10*log10(Tn(idx)));
>> > figure(2);
>> > semilogx(f(idx),10*log10(Fn(idx)));
>> >
>> > If I am interpreting audiowrite correctly, it expects a max absolute
>> > value of unity and your signal is much greater than that.  I think you are
>> > clipping the snot out of the signal in the audio file.
>>
>>
>> Hi Robert, your graph have the same problem than mine.
>> The attenuation stops in low frequency and keep limited.
>> Please, take a look: https://ibb.co/h20qVR
>> Regarding the clipping. Yes, it´s true, it´s clipping, but I need to
>> solve the problem of the attenuation on low frequencies first.
>
>
> Looking at figure 1 from Robert's plot, it seems to me that the pink noise
> input has a steeper slope at low frequency and isn't following the 1/f
> pattern for low frequency.  This corresponds to where the slope deviates
> from the expected.

Hi Ozzy, the input signal is OK. It follows the 1/f pattern.
Thanks and have a good day!
Renato