Friends, In the field of switching power supplies or other power supply architectures in which a switching signal is used, and a signal (output voltage) is sampled and used for control, the prevailing wisdom is that the ADC sample clock should be synchronized so that it samples "between" the switching clock edges and thus avoids coupling the switching noise into the samples. I say this is specious. Sampling theory tells us that, no matter where we sample, the entire waveform can be reconstructed, so regardless of the sampling phase, the same amount of noise energy is coupled into the signal. Can someone reconcile these two points of view? -- Randy Yates % "Watching all the days go by... Digital Signal Labs % Who are you and who am I?" mailto://yates@ieee.org % 'Mission (A World Record)', http://www.digitalsignallabs.com % *A New World Record*, ELO
Sampling "between the noise"
Started by ●June 7, 2011
Reply by ●June 7, 20112011-06-07
The idea of sampling "between" the noise only makes sense if it is done synchronously at the rate of the noise or a sub-harmonic. In that case Nyquist tells us that the noise is aliased to DC and given the proper phase, the "between" part, it turns into a DC component of 0. Can you give a little more info on where you heard of this and what the context is? The devil is in the details. Rick
Reply by ●June 7, 20112011-06-07
On 06/07/2011 12:25 PM, rickman wrote:> The idea of sampling "between" the noise only makes sense if it is > done synchronously at the rate of the noise or a sub-harmonic. In > that case Nyquist tells us that the noise is aliasedSo the scenario you describe is one in which the analog signal is not properly filtered?> to DC and given the proper phase, the "between" part, it turns into > a DC component of 0.I can see it being at DC, but I don't see how it would be 0.> Can you give a little more info on where you heard of this and what > the context is? The devil is in the details.A guy at work. He probably only had a vague idea of what he was talking about (heard it from someone else) because he's not a digital/DSP guy (magnetics). But even in the case you bring up, I don't see the benefit, unless DC is not important (and usually it is in controls). -- Randy Yates % "Watching all the days go by... Digital Signal Labs % Who are you and who am I?" mailto://yates@ieee.org % 'Mission (A World Record)', http://www.digitalsignallabs.com % *A New World Record*, ELO
Reply by ●June 7, 20112011-06-07
PS: I also read about it somewhere, but I can't recall where. It seems to be a common practice. --Randy On 06/07/2011 12:34 PM, Randy Yates wrote:> On 06/07/2011 12:25 PM, rickman wrote: > >> The idea of sampling "between" the noise only makes sense if it is >> done synchronously at the rate of the noise or a sub-harmonic. In >> that case Nyquist tells us that the noise is aliased > > So the scenario you describe is one in which the analog signal > is not properly filtered? > >> to DC and given the proper phase, the "between" part, it turns into >> a DC component of 0. > > I can see it being at DC, but I don't see how it would be 0. > >> Can you give a little more info on where you heard of this and what >> the context is? The devil is in the details. > > A guy at work. He probably only had a vague idea of what he was talking > about (heard it from someone else) because he's not a digital/DSP guy > (magnetics). > > But even in the case you bring up, I don't see the benefit, unless DC > is not important (and usually it is in controls).-- Randy Yates % "Watching all the days go by... Digital Signal Labs % Who are you and who am I?" mailto://yates@ieee.org % 'Mission (A World Record)', http://www.digitalsignallabs.com % *A New World Record*, ELO
Reply by ●June 7, 20112011-06-07
On Tuesday, June 7, 2011 12:34:12 PM UTC-4, Randy Yates wrote:> On 06/07/2011 12:25 PM, rickman wrote: > > > The idea of sampling "between" the noise only makes sense if it is > > done synchronously at the rate of the noise or a sub-harmonic. In > > that case Nyquist tells us that the noise is aliased > > So the scenario you describe is one in which the analog signal > is not properly filtered?I don't understand, oh, you mean anti-alias filtered. Not really. I am assuming that you have a switching PSU and noise is getting into the signal chain rather than coming in with the signal. If the noise is in the signal bandwidth there is nothing you can do with sampling.> > to DC and given the proper phase, the "between" part, it turns into > > a DC component of 0. > > I can see it being at DC, but I don't see how it would be 0.Draw a sine wave and sample it at 90 degrees phase, once per cycle. Then sample it at 0 or 180 degrees phase again once per cycle. Big difference. Since switching noise is mostly spikes at the switch transitions you can sample between them effectively getting a zero contribution from the spikes.> > Can you give a little more info on where you heard of this and what > > the context is? The devil is in the details. > > A guy at work. He probably only had a vague idea of what he was talking > about (heard it from someone else) because he's not a digital/DSP guy > (magnetics). > > But even in the case you bring up, I don't see the benefit, unless DC > is not important (and usually it is in controls).Hmmm... does it make any more sense now? Rick
Reply by ●June 7, 20112011-06-07
On Jun 7, 12:22�pm, Randy Yates <ya...@ieee.org> wrote:> Friends, > > In the field of switching power supplies or other power supply > architectures in which a switching signal is used, and a signal > (output voltage) is sampled and used for control, the prevailing > wisdom is that the ADC sample clock should be synchronized so that it > samples "between" the switching clock edges and thus avoids coupling > the switching noise into the samples. > > I say this is specious. Sampling theory tells us that, no matter where > we sample, the entire waveform can be reconstructed, so regardless of > the sampling phase, the same amount of noise energy is coupled into > the signal. > > Can someone reconcile these two points of view? > -- > Randy Yates � � � � � � � � � � �% "Watching all the days go by... > Digital Signal Labs � � � � � � �% �Who are you and who am I?" > mailto://ya...@ieee.org � � � � �% 'Mission (A World Record)',http://www.digitalsignallabs.com% *A New World Record*, ELOYou're making the assumption that you're sampling the noisy signal fast enough to reconstruct it from its samples, which isn't necessarily true. Plus, I would probably term the coupling from the switching signal "interference" instead of "noise." In a white noise background, yes, you'll have the same amount of noise energy in the sampled signal no matter the sampling phase. Coupling from the supply's switching signal isn't white noise at all, however: the coupling effect may be predictable in time and in shape. I don't see any reason why you couldn't take advantage of this knowledge to obtain samples that were clear of the interference. I'm not a power supply designer, but I don't know of any requirement to sample the output voltage fast enough such that the Nyquist criterion is met for the switching interference that couples to the output. Jason
Reply by ●June 7, 20112011-06-07
On 06/07/2011 12:41 PM, rickman wrote:> On Tuesday, June 7, 2011 12:34:12 PM UTC-4, Randy Yates wrote: >> On 06/07/2011 12:25 PM, rickman wrote: >> >>> The idea of sampling "between" the noise only makes sense if it is >>> done synchronously at the rate of the noise or a sub-harmonic. In >>> that case Nyquist tells us that the noise is aliased >> >> So the scenario you describe is one in which the analog signal >> is not properly filtered?> I don't understand, oh, you mean anti-alias filtered.Yes.> Not really. I am assuming that you have a switching PSU and noise is > getting into the signal chain rather than coming in with the signal.So, for example, coupling into the trace between the filter and the ADC input terminal? I see this is a different input mechanism than via filter, but as far as the ADC is concerned, it's out-of-band energy.> If the noise is in the signal bandwidth there is nothing you can do > with sampling.> >>> to DC and given the proper phase, the "between" part, it turns into >>> a DC component of 0. >> >> I can see it being at DC, but I don't see how it would be 0. > > Draw a sine wave and sample it at 90 degrees phase, once per cycle. > Then sample it at 0 or 180 degrees phase again once per cycle. Big > difference. Since switching noise is mostly spikes at the switch > transitions you can sample between them effectively getting a zero > contribution from the spikes.Yeah, for a sine wave. We weren't talking about a sine wave but _noise_. But I see now that you're saying that we can null out the _fundamental_ of noise with a periodic component, which I can see could be beneficial since the fundamental usually carries the most energy.>>> Can you give a little more info on where you heard of this and what >>> the context is? The devil is in the details. >> >> A guy at work. He probably only had a vague idea of what he was talking >> about (heard it from someone else) because he's not a digital/DSP guy >> (magnetics). >> >> But even in the case you bring up, I don't see the benefit, unless DC >> is not important (and usually it is in controls). > > Hmmm... does it make any more sense now?After clarifying what you meant, yes. Thanks Rick. -- Randy Yates % "Watching all the days go by... Digital Signal Labs % Who are you and who am I?" mailto://yates@ieee.org % 'Mission (A World Record)', http://www.digitalsignallabs.com % *A New World Record*, ELO
Reply by ●June 7, 20112011-06-07
On 06/07/2011 12:50 PM, Jason wrote:> On Jun 7, 12:22 pm, Randy Yates<ya...@ieee.org> wrote: >> Friends, >> >> In the field of switching power supplies or other power supply >> architectures in which a switching signal is used, and a signal >> (output voltage) is sampled and used for control, the prevailing >> wisdom is that the ADC sample clock should be synchronized so that it >> samples "between" the switching clock edges and thus avoids coupling >> the switching noise into the samples. >> >> I say this is specious. Sampling theory tells us that, no matter where >> we sample, the entire waveform can be reconstructed, so regardless of >> the sampling phase, the same amount of noise energy is coupled into >> the signal. >> >> Can someone reconcile these two points of view? >> -- >> Randy Yates % "Watching all the days go by... >> Digital Signal Labs % Who are you and who am I?" >> mailto://ya...@ieee.org % 'Mission (A World Record)',http://www.digitalsignallabs.com% *A New World Record*, ELO > > You're making the assumption that you're sampling the noisy signal > fast enough to reconstruct it from its samples, which isn't > necessarily true. Plus, I would probably term the coupling from the > switching signal "interference" instead of "noise." In a white noise > background, yes, you'll have the same amount of noise energy in the > sampled signal no matter the sampling phase. Coupling from the > supply's switching signal isn't white noise at all, however: the > coupling effect may be predictable in time and in shape. I don't see > any reason why you couldn't take advantage of this knowledge to obtain > samples that were clear of the interference. I'm not a power supply > designer, but I don't know of any requirement to sample the output > voltage fast enough such that the Nyquist criterion is met for the > switching interference that couples to the output.That's essentially what Rick said. Thanks Jason. -- Randy Yates % "Watching all the days go by... Digital Signal Labs % Who are you and who am I?" mailto://yates@ieee.org % 'Mission (A World Record)', http://www.digitalsignallabs.com % *A New World Record*, ELO
Reply by ●June 7, 20112011-06-07
On Jun 7, 12:22�pm, Randy Yates <ya...@ieee.org> wrote:> Friends, > > In the field of switching power supplies or other power supply > architectures in which a switching signal is used, and a signal > (output voltage) is sampled and used for control, the prevailing > wisdom is that the ADC sample clock should be synchronized so that it > samples "between" the switching clock edges and thus avoids coupling > the switching noise into the samples. > > I say this is specious. Sampling theory tells us that, no matter where > we sample, the entire waveform can be reconstructed, so regardless of > the sampling phase, the same amount of noise energy is coupled into > the signal. > > Can someone reconcile these two points of view? > -- > Randy Yates � � � � � � � � � � �% "Watching all the days go by... > Digital Signal Labs � � � � � � �% �Who are you and who am I?" > mailto://ya...@ieee.org � � � � �% 'Mission (A World Record)',http://www.digitalsignallabs.com% *A New World Record*, ELOyea we do this all the time, sampling theory goes out the window when you have prior information about the signal, in this case when the spikes are occurring, it's used because we have noisy but small and efficient switching power supplies but need to convert low voltage signals
Reply by ●June 7, 20112011-06-07
On 06/07/2011 09:22 AM, Randy Yates wrote:> Friends, > > In the field of switching power supplies or other power supply > architectures in which a switching signal is used, and a signal > (output voltage) is sampled and used for control, the prevailing > wisdom is that the ADC sample clock should be synchronized so that it > samples "between" the switching clock edges and thus avoids coupling > the switching noise into the samples. > > I say this is specious. Sampling theory tells us that, no matter where > we sample, the entire waveform can be reconstructed, so regardless of > the sampling phase, the same amount of noise energy is coupled into > the signal. > > Can someone reconcile these two points of view?Sampling theory tells us that _if_ the signal in question is of limited bandwidth compared to the sampling rate, then no matter where we sample the entire waveform can be reconstructed. A quick look at the voltages on a switched amplifier tells us that there are humongous spikes that are very narrow compared with the switching rates. This, in turn, tells us that there is considerable energy in the spectrum above the Nyquist rate, if we were to sample at the switching frequency. That same quick look, however, tells us that we can toss sampling theory out the window and use common sense: we see relatively clean voltages, punctuated by those humongous spikes. So we know that if we can cleanly sample those voltages when they are at their clean periods, then we can get much better data than if we were to let those humongous spikes pollute our measurements. -- 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
