Forums

testing a digital filterbank

Started by rameshk December 5, 2010
Hi All,

I tried browsing topics on filterbanks on this site so as not to repeat my
question. But haven't found any posts that is similar to my problem.

I have a 16-channel digital filterbank based on polyphase decomposition. My
objective is to find the absolute delay a signal undergoes in one of these
filters. 

To test this, a 10-us square pulse was amplitude modulated by a 50 MHz
carrier and fed to the input of the filterbank. The filterbank was run
clocked at 320 MHz which in turn gives me 16 times 10 MHz subbands, each
band tuned to 0, 10, 20, ... 150 MHz.

Now I expect my square pulse to appear in the subband centred at 50 MHz.
Since the filterbank generates  complex outputs, I simply computed the
absolute value of this signal expecting to see the square pulse in the data
stream. However I do not see any such pulse in the filterbank output.

My questions are:
1.Do I have to demodulate the signal in order to get back the square
pulse?
2. If it is AM demodulation, how do I generate the I/Q's - the in phase and
quadrature components? By computing the product of the filterbank output
samples and  cos(w_c*t) and sin(w_c*t) , where w_c is the angular carrier
frequency?

Thanks in advance for your responses.

Cheers,
Ramesh






On 12/05/2010 05:37 AM, rameshk wrote:
> Hi All, > > I tried browsing topics on filterbanks on this site so as not to repeat my > question. But haven't found any posts that is similar to my problem. > > I have a 16-channel digital filterbank based on polyphase decomposition. My > objective is to find the absolute delay a signal undergoes in one of these > filters. > > To test this, a 10-us square pulse was amplitude modulated by a 50 MHz > carrier and fed to the input of the filterbank. The filterbank was run > clocked at 320 MHz which in turn gives me 16 times 10 MHz subbands, each > band tuned to 0, 10, 20, ... 150 MHz. > > Now I expect my square pulse to appear in the subband centred at 50 MHz. > Since the filterbank generates complex outputs, I simply computed the > absolute value of this signal expecting to see the square pulse in the data > stream. However I do not see any such pulse in the filterbank output. > > My questions are: > 1.Do I have to demodulate the signal in order to get back the square > pulse? > 2. If it is AM demodulation, how do I generate the I/Q's - the in phase and > quadrature components? By computing the product of the filterbank output > samples and cos(w_c*t) and sin(w_c*t) , where w_c is the angular carrier > frequency?
If you notice that this question hasn't been answered yet it's probably because people are wondering where to start. I know what our resident curmudgeon would say... 1: What do you think? If you take your unfiltered pulse, take it's absolute value, and look at the results, do you get a square pulse? If not, why do you expect to get a square pulse out of a filtered version? 1a: How narrow are these filters? Taking the Fourier transform of that question, how much is the 50MHz filter going to spread your narrow (and harmonic rich) pulse? 2: What do you mean? What do you expect to mean? For that matter, what is the mapping between the real input of your filters (I assume they're real, you didn't mention anything else) and the complex output? -- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" was written for you. See details at http://www.wescottdesign.com/actfes/actfes.html
Hi Tim,
Thanks for your reply.

>If you notice that this question hasn't been answered yet it's probably >because people are wondering where to start.
yeah, I agree that the question isn't well-posed.
>I know what our resident curmudgeon would say... > >1: What do you think? If you take your unfiltered pulse, take it's >absolute value, and look at the results, do you get a square pulse? If >not, why do you expect to get a square pulse out of a filtered version?
Yep, I do expect to see a square pulse when I take the absolute value. Filtered version: depends on how wide the filter is. The pulse will be convolved by the filter's impulse response.
>1a: How narrow are these filters? Taking the Fourier transform of that >question, how much is the 50MHz filter going to spread your narrow (and >harmonic rich) pulse?
As I wrote above, the filters are 10 MHz wide, and my pulse is only 10 microseconds wide. My thinking was, this pulse would have a sin(x)/x frequency response, with the first null at 1/(0.1 us), or 100 KHz. And I justified that a 10-MHz filter will not affect the pulse much.
>2: What do you mean? What do you expect to mean? For that matter, what >is the mapping between the real input of your filters (I assume they're >real, you didn't mention anything else) and the complex output?
Sorry, I should mentioned that these are complex filters. The 16 bandpass filters in question are all complex bandpass filters that generates complex output. Again, the objective of modulating a square pulse using the 50 MHz carrier was to make sure that only the filter tuned to 50 MHz will shows the pulse. But now I am stuck at trying to demodulate the 50 MHz carrier to retrieve the square pulse. Thanks again for your responses. Cheers, Ramesh