Thomas From your description, it looks like you want to test a 3rd order comb (decimation??- why decimation?) filter using a test signal with flat spectrum. If you want to test a filter, why would you choose a signal which would have the characteristics as that of noise. I am unable to understand. Normally a sweep signal is chosen to observe the frequency response of a filter. Please give exact context of the problem. Do let me know. Regards arun -----Original Message----- From: Jeff Brower [mailto:] Sent: Wednesday, January 26, 2005 7:26 PM To: Thomas Patrick Cc: Subject: Re: [matlab] Digital Filter Test with LFSR Thomas- > It is a known fact that LFSR generate flat spectrum, I choosed > LFSR to test my 3rd order polyphase comb decimation filter with > decimation factor of 256. I am wondering what would be the > lenght of LFSR ? it should be 4 bit or 8 bit or 10 bit ?? All > generate flat spectrum. What characteristics help me in choosing > the lenght of LFSR ??. > > I have read few technical papers, In which LFSR is being used > in Build In Self Test(BIST) to test Sigma delta modulator and > no where specified the lenght of LFSR. > > I have one idea to select to lenght. > > 1) Its depends on the how many pattern we would like to generate > for example 8 bit LFSR gives 2^8 -1 and 10 bit generate 2^10-1. > More pattern would give us better results. > > 2) The sequence should be longer than the delay response of filter. Expected system impulse response length is one good criteria. As one example, some acoustic measurement systems use a MLS method (maximum length sequence) which is similar to LFSR. The length of non-repeating pseudo-random sequence is chosen based on the expected measurement situation. For instance, in measuring the response of a concert hall with lots of reflections and echos it's necessary to choose a long sequence -- 18 bit or more (equivalent to 5 sec at 48 kHz sampling rate). -Jeff |