Comments on: Polyphase filters for interpolation

By: Krishna Sankar

Krishna Sankar — Mon, 12 Oct 2009 00:36:14 +0000

@Talib: It really does not matter, no? length(y1-y) and length(y1) reports the same number. Agree?

By: Krishna Sankar

Krishna Sankar — Mon, 12 Oct 2009 00:34:27 +0000

@Talib: My replies:
1/ In the code, I assumed a sampling rate of 1kHz
2/ L is the oversampling factor. I used a small matlab code snippet to plot the frequency response
octave:12> L = 4
octave:13> h = sinc([-20:20]/L);
octave:14> hF = fft(h,1024);
octave:15> plot([-512:511]/1024,(abs(fftshift(hF))));
octave:16> xlabel(‘freq, kHz’); ylabel(‘amplitude’);

Hope this helps.

By: Talib

Talib — Thu, 08 Oct 2009 18:59:12 +0000

Also Krishna,

Shouldnt:

diff = (y1-y)*(y1-y)’/length(y1-y)

actually be…

diff = (y1-y)*(y1-y)’/length(y1) ?

Thanks again,

By: Talib

Talib — Thu, 08 Oct 2009 17:59:14 +0000

Hi Krishna,

Great article on polyphase – question though – could you go through in more detail about EXACTLY how you made this sinc-filter? What is its sampling rate?… Why is ‘L’ an argument? … I understand that you need to low pass filter, and that a sinc in time is a rectangle in frequency, but exactly how did you select its arguments?…

Thanks!

By: Krishna Pillai

Krishna Pillai — Mon, 06 Jul 2009 14:25:01 +0000

@saira: No special reason, I just wanted to define a filter which provided around 40dB attenuation outside the passband.

By: Saria

Saria — Sun, 05 Jul 2009 14:33:20 +0000

Dear Mr. Krishna, thank you for the wonderful explanation of polyphase filters. I would like to know how you chose the sinc filter time range [-20:20]/L,which will give a vector of 41 values in time domain?

By: Krishna Pillai

Krishna Pillai — Sun, 21 Jun 2009 07:20:27 +0000

@sumo: Nice question. Let me try to answer…
With the original sampling frequency of 1000Hz, we were able to ’see’ frequencies from [-500Hz to +500Hz). Now, with the oversampled frequency of 4000Hz, we now can ’see’ frequencies from [-2000Hz to 2000Hz). Now as the specrum gets replicated at multiples for sampling frequency, the frequency at 200Hz is replicated at 1200Hz, -800Hz, -1800Hz. So, instead of 1 frequency, we have four frequencies at 1/4th amplitude. Hence the reduction in magnitude by 20*log10(1/4) = -12dB.

Do you agree?