DC offset removal through FPGA
in [DSP]
|
Prev: SHARC debug
Next: Who Invented the Z Transform
From: Rick Lyons on 29 Jun 2005 09:08 On Sun, 26 Jun 2005 09:55:58 -0400, Jerry Avins <jya(a)ieee.org> wrote: > >Why is direct superior to going around the other side of the barn? If >the final result is the same, the simplest approach is usually to be >preferred. Are there hidden assumptions we don't share? > >Jerry Hi, This thread interests me because I don't know any simple way to remove the DC portion of a signal in real time. It seems to me that any digital IIR filter method would induce some sort of unpleasant phase nonlinearity. It sounds like the "subtract the output of a super-narrow (linear phase) lowpass filter from the original signal" method should work. I wonder if it does. I wish I understood Matt Timmerman's method. [-Rick-]
From: Sandro on 29 Jun 2005 10:16 Rick Lyons wrote: > This thread interests me because I don't know any > simple way to remove the DC portion of a signal in real > time. It seems to me that any digital IIR filter > method would induce some sort of unpleasant phase > nonlinearity. > > It sounds like the "subtract the output of a > super-narrow (linear phase) lowpass filter from the > original signal" method should work. > I wonder if it does. > > I wish I understood Matt Timmerman's method. Hi, Here's a nice xilinx document (TechXclusives) about removing DC offset... Digitally Removing a DC Offset (or "DSP Without Math?") Note: Part 1: not Xilinx FPGA specific ;-) Part 2: how eficiently implement on Xilinx FPGAs http://www.xilinx.com/xlnx/xweb/xil_tx_display.jsp?sGlobalNavPick=&sSecondaryNavPick=&category=&iLanguageID=1&multPartNum=1&sTechX_ID=kc_dig_offset Sandro
From: Steve Underwood on 29 Jun 2005 10:43 Rick Lyons wrote: > On Sun, 26 Jun 2005 09:55:58 -0400, Jerry Avins <jya(a)ieee.org> wrote: > > > >>Why is direct superior to going around the other side of the barn? If >>the final result is the same, the simplest approach is usually to be >>preferred. Are there hidden assumptions we don't share? >> >>Jerry > > > Hi, > This thread interests me because I don't know any > simple way to remove the DC portion of a signal in real > time. It seems to me that any digital IIR filter > method would induce some sort of unpleasant phase > nonlinearity. > > It sounds like the "subtract the output of a > super-narrow (linear phase) lowpass filter from the > original signal" method should work. > I wonder if it does. How is "phase linear" of relevance in a signal so narrowly focussed around DC? I find DC rather hard to phase shift. Regards, Steve
From: Jerry Avins on 29 Jun 2005 11:41 Rick Lyons wrote: > On Sun, 26 Jun 2005 09:55:58 -0400, Jerry Avins <jya(a)ieee.org> wrote: > > > >>Why is direct superior to going around the other side of the barn? If >>the final result is the same, the simplest approach is usually to be >>preferred. Are there hidden assumptions we don't share? >> >>Jerry > > > Hi, > This thread interests me because I don't know any > simple way to remove the DC portion of a signal in real > time. It seems to me that any digital IIR filter > method would induce some sort of unpleasant phase > nonlinearity. > > It sounds like the "subtract the output of a > super-narrow (linear phase) lowpass filter from the > original signal" method should work. > I wonder if it does. > > I wish I understood Matt Timmerman's method. Since it takes a time T ~= 1/f to isolate a particular frequency, any DC measurement necessarily comes too late to be useful. :-) Although we treat offsets due to drift as DC, we could null them out once for all time if they really were. What we can do in practice is estimate the very-low-frequency departure from zero amplitude often enough to keep it within bounds by applying a correction, or remove those frequencies whose amplitudes we null out from the stream of delivered data. I had said that the two methods were equivalent. I was wrong. They certainly seem to be equivalent: consider an odd-tap FIR low-pass with a cut-off of, say, half a Hertz. It's a long filter, with lots of delay. Its output, subtracted from the delayed original signal is in fact identical from the output of its high-pass sister constructed by inverting the sign of all its taps, then adding [the number] one to the middle tap. Another way to look at the high-pass sister is as an efficient way to implement lowpass-and-subtract; the equivalence is complete. Unfortunately, the method in either guise has too much delay. Recognizing that "DC" *THEN* is very nearly "DC" *NOW* and subtracting "DC" *THEN* from the undelayed (or only slightly delayed) signal allows us to exploit the slow nature of the DC (and environs) to be removed without incurring much delay. We're always playing catchup, but we're never too far wrong. Servo designers take note: *Avins's Method* (ahem!) trades away a little bit of accuracy and gets a lot of promptness in return. Since the offset needs be updated only at a rate suitable for sampling those low frequencies, not much effort needs to be spent to calculate it. CAZ anyone? Jerry -- Engineering is the art of making what you want from things you can get. ýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýý
From: robert bristow-johnson on 29 Jun 2005 11:38
in article d9uc20$3s8$1(a)nnews.pacific.net.hk, Steve Underwood at steveu(a)dis.org wrote on 06/29/2005 10:43: > Rick Lyons wrote: > >> On Sun, 26 Jun 2005 09:55:58 -0400, Jerry Avins <jya(a)ieee.org> wrote: >> >>> Why is direct superior to going around the other side of the barn? If >>> the final result is the same, the simplest approach is usually to be >>> preferred. Are there hidden assumptions we don't share? >>> >> This thread interests me because I don't know any >> simple way to remove the DC portion of a signal in real >> time. It seems to me that any digital IIR filter >> method would induce some sort of unpleasant phase >> nonlinearity. it's costly, but simple, but a long phase linear FIR filter to get the "DC" to subtract from the appropriately delayed signal has no phase non-linearity. >> It sounds like the "subtract the output of a >> super-narrow (linear phase) lowpass filter from the >> original signal" method should work. and it's a HPF unless you put something non-linear, like a median filter, on that "DC" output of a LPF before subtracting. (i wonder how that might sound?) > How is "phase linear" of relevance in a signal so narrowly focussed > around DC? I find DC rather hard to phase shift. there's the transition band. it's funny, on one hand there are people saying we can't hear phase at all (see other thread about it) and those who say that the phase non-linearity of a DC blocking HPF (hell, we can put the cutoff at 0.1 Hz, the filter will be pretty flattened out by 20 Hz). both extremes are extreme. -- r b-j rbj(a)audioimagination.com "Imagination is more important than knowledge." |