A Useful Source of Signal Processing Information

Hello Sir, can you suggest me a method to design very narrow band window. For example with 70MHz sampling frequency, can we design a window which as +/-500kHz main lobe width with -20 dB side lobe suppression. I found Chebyshev window is good, but can we find a method which meets my specs?. This is basically to remove the tone of order 800kHz, in a 70MHz sampling rate system. Please suggest me on this.
Thank you

Hello kovelapudi,
When you use the word "window" I think of time-domain window sequences such as Hanning, Hamming, Blackman, etc. You say you want to remove an 800 kHz tone from some signal, which sounds like you want to pass your signal through a filter that has a notch located at 800 kHz. But window sequences are not used for filtering. So I'm confused. Can you tell me exactly what you mean by your word "window"? Are you trying to design a time-domain window sequence or are you trying to design a digital notch filter?

Hello sir,
Yes, I want to design a digital filter which should filter 800 kHz tone (i.e. suppress the rest of the signal except 800 kHz tone, we can say like this a low pass filter with 800 kHz pass band ). Here window refer to window sequence only (Hanning, Hamming, Kaiser e.t.c). My requirements are "Sampling rate of the system 70 MHz" and I want to use minimum number of taps (48 or 50 e.t.c) for the filter, and I what to have -25 dB side lobe suppression. Initially I tried with 48 coefficients Chebyshev window itself as filter, with 40 dB side lobe suppression, but with this 70 MHz sampling rate, I am unable to get pass band width less than +/-2.5MHz, but as I explained I need pass band less than this (800 kHz,(+/-400 kHz)). Is there any other window sequence of 48 coefficients which gives narrow pass band (800 kHz) at this sampling rate or any other methods to achieve this with minimum coefficients. Is there any criteria on minimum pass band of filter in relation to the sampling frequency.

Hello kovelapudi,
I think can help you if I could just figure out what you are doing. You insist on using the word "window." In the world of DSP a "window" is a sequence of numbers (sometimes called a "window sequence"). And in the world of DSP "windows" are used for spectrum analysis, NOT digital filtering. So kovelapudi, as clearly and carefully as you can please tell me what you mean when you use the word "window." That is, what EXACTLY does the word "window" mean to you.

Hello sir,
Sorry for the uncertain information. Here window means window sequence only like hanning, hamming window sequences in the DSP world as you identified. As you said, window techniques applies on filter coefficients for spectral analysis. I have observed frequency response plots of lowpass filter (with 70 MHz sampling rate,800 kHz cutoff frequency) with hanning window, hamming and Chebyshev windows using "Matlab fdatool " with filter order set to "48" (48 taps low pass filter). what I observed with Chebyshev window with 40dB side lobe suppression is wider pass band i.e. "I am unable to get pass band less than +/-1 MHz" , even though I gave 0.8MHz as cutoff freq. So I need a method which helps me to achieve very narrow band low pass filter (pass band <1 MHz), with minimum coefficients (48 taps here ) and minimum 25dB side lobe suppression, with 70MHz sampling rate. I used the same values in matlab "fdatooll" (filter design and analysis tool of matlab) which I presented here.
Thank you

Hello kovelapudi,
I did not say "...window techniques applies on filter coefficients for spectral analysis." I said just the opposite! I said windows are NOT used for digital filtering. For your filter design problem make the idea of windows leave your head. Do NOT think about windows in any way for your filter design problem. Also, I suggest you stop using the 'fdatool'.

Do the following steps shown by '[ ]':
[1] Run the following code:

Num_Taps = 200; % Number of coefficients
B = firpm(Num_Taps,[0,0.8/(70/2),2.2/(70/2),1],[1,1,0,0]);

Freq_Resp = fft(B, 1024);
Freq_Mag_Resp = abs(Freq_Resp);
Freq_Mag_Resp_dB = 20*log10(Freq_Mag_Resp/max(Freq_Mag_Resp));
Freq = linspace(0, 70*(1023)/1024, 1024);

figure(1)
plot(Freq, Freq_Mag_Resp_dB)
axis([0, 70, -100, 5])
grid on, zoom on, xlabel('MHz'), ylabel('dB')

[2] Zoom & examine the plot near 0.8 MHz.
[3] Examine the stopband attenuation level.
[4] Next, set Num_Taps = 150 & run the code.
[2] Zoom & examine the plot near 0.8 MHz.
[3] Examine the stopband attenuation level.
[4] Set Num_Taps = 100 & run the code.
[2] Zoom & examine the plot near 0.8 MHz.
[3] Examine the stopband attenuation level.
[4] Set Num_Taps = 50 & run the code.
[2] Zoom & examine the plot near 0.8 MHz.
[3] Examine the stopband attenuation level.

You will see that a 50-tap FIR filter cannot satisfy your desired frequency response. kovelapudi, your desired filter's passband width is such a small number relative to the 70 MHz sample that you will need a FIR filter with hundreds of taps. If hundreds of taps is not possible for you, then your only hope is, (1) use decimating cascaded integrator-comb filters followed by a FIR lowpass filter, (2) use a cascade of decimating half-band FIR filters followed by a FIR filter, or (3) use an "interpolated FIR" (IFIR) filter. Any of those three choices me means you have MUCH study & work to do. Who gave you such a difficult filter design problem(!)?

Hello sir,
Thanks for your suggestions . I have simulated as suggested. Now I need to take any one of the available choices.

Thank you.

Sir,

Please help me to create a bessel filter of 3Hz cutoff at 300 samples per second from ADC "AD7190" . Which is needed to interface a load cell to MCU?