PLL for Zero Delay LPF
in [Design]
|
From: Vladimir Vassilevsky on 23 Jun 2010 19:18 Glenn Kenroy wrote: > On Wed, 23 Jun 2010 08:12:49 -0700, John Larkin > <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: > > > >>Does your filter really need to be realtime? If you're acquiring and >>digitizing the data, it can be post-processed, which allows a >>near-ideal lowpass to be applied. >> >>A zero delay realtime lowpass filter is impossible. >> > > > OK, but since the primary objective is to remove 60Hz artifact from > the ELF geomagnetic data (in real time), would a zero delay _notch_ > filter be feasible in place of a LPF? Yes. It should be infinitely narrow notch. Does it make you feel better? Get real. Set up the practical requirements for your project avoiding zeroes and infinities. VLV
From: Dave Platt on 23 Jun 2010 19:58 In article <4c2287e0.662703(a)news.tpg.com.au>, Glenn Kenroy <glennkenroy(a)protech.com> wrote: > >>Does your filter really need to be realtime? If you're acquiring and >>digitizing the data, it can be post-processed, which allows a >>near-ideal lowpass to be applied. >>A zero delay realtime lowpass filter is impossible. >OK, but since the primary objective is to remove 60Hz artifact from >the ELF geomagnetic data (in real time), would a zero delay _notch_ >filter be feasible in place of a LPF? The same problem exists. Fundamentally (>>grin<<), what you're looking to do here is impossible - at least, if you're looking for anything close to a general-case solution to the problem. If you want to think about the issue in a verbal (and thus sloppy way), consider it this way: there is an inextricable relationship between "frequency" and "time". At any one given instant, you can't legitimately say that a signal has a particular frequency - you can only draw conclusions about frequency after you've observed the signal for a period of time. That holds for you, me, and any circuit or DSP system that you want to put together. The narrower you want your frequency analysis to be, the longer the period of time over which you have to study the signal before you can make a "pass it" or "reject it" decision. You cannot, for example, distinguish with accuracy between 60 Hz and 59 Hz in a matter of a millisecond or so - you haven't seen enough of the signal to do so. Again, that applies to people, electronics, and computers. You can't filter a signal without delaying it, in one way or another. Now, there's a very specific special-case subset of the question you're asking about, for which a filter is possible - with a very significant limitation. Specifically: if you want to notch out a signal component (such as 60 Hz), *and* you know that its amplitude isn't changing over time, then you can play some tricks. You can split the signal in half, and feed part of it to a very narrow-bandwidth "band-pass" filter which lets through only the 60 Hz component, You design this band-pass filter carefully, so that its amplitude exactly matches that of the 60 Hz component which is coming in, and its total time delay is an exact multiple of 1/60 of a second. In effect, this bandpass filter will "track" the 60 Hz component of the incoming signal, quite accurately, in both amplitude and phase. You can now subtract the bandpassed 60 Hz interference signal, from the other copy of your incoming signal. This will act as a narrow notch filter, canceling out the 60 Hz component while letting the other portions of the signal pass through with no delay. Seems like what you want, right? Well... maybe. As I said, it has a significant limitation. It will filter out the 60 Hz signal *only* if its amplitude doesn't change at all. Every time the amount of 60 Hz interference in your incoming signal changes, it will take the bandpass filter a brief moment to "catch up" with the change. The narrower you made it, the longer it will take to catch up. Until it does, it will be injecting the wrong ("outdated") amount of cancellation into your signal mixer... it'll either under-cancel or over-cancel the actual amount of interference at that moment. In either case, you'll get a "burble" of 60 Hz noise in your signal, until the bandpass filter catches up with reality... and, inevitably, this takes time. -- Dave Platt <dplatt(a)radagast.org> AE6EO Friends of Jade Warrior home page: http://www.radagast.org/jade-warrior I do _not_ wish to receive unsolicited commercial email, and I will boycott any company which has the gall to send me such ads!
From: Joel Koltner on 23 Jun 2010 20:20 "Bill Sloman" <bill.sloman(a)ieee.org> wrote in message news:8caedc1c-2704-4ce9-8505-692d6d020f6f(a)i31g2000yqm.googlegroups.com... On Jun 23, 5:16 pm, Joerg <inva...(a)invalid.invalid> wrote: >http://en.wikipedia.org/wiki/Thiotimoline >It is - of course - too good to be true. I can build you negative group delay filters. Would that be close enough? :-)
From: Sjouke Burry on 23 Jun 2010 22:26 Glenn Kenroy wrote: > On Wed, 23 Jun 2010 08:12:49 -0700, John Larkin > <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: > > >> Does your filter really need to be realtime? If you're acquiring and >> digitizing the data, it can be post-processed, which allows a >> near-ideal lowpass to be applied. >> >> A zero delay realtime lowpass filter is impossible. >> > > OK, but since the primary objective is to remove 60Hz artifact from > the ELF geomagnetic data (in real time), would a zero delay _notch_ > filter be feasible in place of a LPF? > > Glenn Kenroy Whats wrong with sampling the data using a pll to clean up and sync to the 60 hz ?? That or something like that has been done for ages.
From: Glenn Kenroy on 23 Jun 2010 22:34
On Wed, 23 Jun 2010 16:12:30 -0700, John Larkin <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: >Another approach would be to have a local 60 Hz reference. >Synchronously detect the i/q components in your signal, remodulate, >and subtract them out. > You mean perhaps to use two identical sensing coils; one tuned to ELF and the second to 60Hz? The output from the second could then be adjusted in amplitude and inverted to null the 60Hz component in the output of the second. >Does your definition of "real time" allow any time delay? If so, a >lowpass filter with some delay and phase equalization might work. > Yes, but I am still uncertain how the phase equalization would be implemented in terms of actual circuitry. Glenn Kenroy |