Technical discussions about Freescale (Motorola) DSPs (including the DSP56000, DSP56300, DSP56600, 56800 DSPs).
|
I am looking for opinions about allpass filters; advantages and disadvantages. I am implementing a number of IIR first order lowpass, highpass, and shelving filters. I have done some filters as IIR direct form, and others using the allpass structure. For example, using allpass structure; A(Z) = (-alpha + Z^(-1)) / (1 - alpha * Z^(-1)) H_lpf(Z) = 1/2 * [1 + A(Z)] H_shelving(Z) = 1/2 * [1 + A(Z)] + K/2 * [1 - A(Z)] I am looking for opinions. Why should I use allpass structure instead of direct forms for these first order IIR filters? So far I have: Advantage: Tunable with one, or two parameters (alpha and K), thus I can code up one function; output = ALLPASS_compute(input, alpha) and compute many different filters, using the same code. Disadvantage: The factor of 1/2 means I have to throw a bit away (unless I keep track of double precesion result with the function call), or perhaps its 1/2 bit if I use rounding. Perhaps its more MIPS to compute the filters modularly using an ALLPASS_compute() function, although perhaps less code space. Kevin www.checkeffect.net |
|
|
|
As you
undoubtedly know, IIR filters come in multiple forms.
The DSP
version of an IIR filter than you want to use has the format:
y[n] = A *
y[n-1] + B * x[n] + C*x[n-1]
where x[n] is
the most recent input sample and y[n] is the most recent output sample.
Sometimes,
one or more of the constants may be a 0, eliminating a term. If you are pressed for processing
power, or want to keep power consumption low, I would recommend reducing the solution in your
code rather than at run time. Often, the labeling of the constants varies from developer to
developer based on personal preference and training.
I personally
like Matlab for filter design. You can easily create any filter (butterworth, chebychev, etc)
with a few simple parameters and get all the coefficients. For a first order filter, the family
of the filter is generally not relevant. They also have something called "sptool" that can be
useful for custom filter design.
When using an
IIR filter, in order to keep quantization low, it is very important to use (and store) an
accumlator that is wider than the input signal. Depending on where the zeros and poles
are, not doing this can have little effect, lots of quantization noise, or even make the
entire filter unstable. A "bit true" simulation of the filter can show the imact of not
doing this.
The common
method is to store y[n] (for use as y[n-1] on the next iteration) as a double precisision and
also to use double precision for the accumulation.
Howard
|
|
Hi everybody! do you know if new Dsp56321 could be Jtag tested with same hardware (command converter + Isa card) used with 56309 processors ? Have you tested it ? I like very much to use old emulation system without buy another one :-) Thak you in advance Filippo Icosystel R&D Vazia (RI) ===== -- F.Tigli SCPOP author --- Visit our Web site www.scpop.de --------------------------- _______________________________ |
|
|
|
The old UCC (Universal Command Converter) as it's called that was comprised of an ISA card and external converter board is very, very outdated as far as software goes. It hasn't been supported by Suite56 for several years. Therefore, you won't be able to set the proper device type and there will be some software bugs that you'll have to deal with. It would be in your best interest to purchase a new Parallel Port Command Converter so you can run the latest software. I've seen them on the market for ~$150 and some similar, yet compatible versions from other manufacturers for even less... -- dB --- Tigli Filippo <> wrote: > Hi everybody! > do you know if new Dsp56321 could be Jtag tested with > same hardware (command converter + Isa card) used with > 56309 processors ? > Have you tested it ? > > I like very much to use old emulation system without > buy another one :-) > > Thak you in advance > > Filippo > Icosystel R&D > Vazia (RI) > > ===== > -- F.Tigli SCPOP author --- > > Visit our Web site > www.scpop.de > > --------------------------- > > > _______________________________ > ">http://mail.yahoo.com |