Empathy List Archives

HM

Hal Murray

Thu, Jan 10, 2019 8:37 PM

Case A: you simply throw away samples, keeping only every nth sample,
without regard for the frequency content of the original signal.

Case B: you first perform appropriate anti-aliasing filtering on the
original signal, and only then throw away all but every nth sample

I don't think case A makes sense. You are throwing away information. You
will get aliasing.

Case B can be more complicated than a traditional low pass anti-aliasing
filter. Consider a simple SDR (software defined radio). The front end ADC
captures a wide bandwidth, a filter selects a narrow chunk that you are
interested in. Decimation is appropriate. The filter threw away the out of
band signal that would get aliased on top of your signal if you didn't filter.

--
These are my opinions. I hate spam.

k8yumdoober@gmail.com said: > Case A: you simply throw away samples, keeping only every nth sample, > without regard for the frequency content of the original signal. > Case B: you first perform appropriate anti-aliasing filtering on the > original signal, and only then throw away all but every nth sample I don't think case A makes sense. You are throwing away information. You will get aliasing. Case B can be more complicated than a traditional low pass anti-aliasing filter. Consider a simple SDR (software defined radio). The front end ADC captures a wide bandwidth, a filter selects a narrow chunk that you are interested in. Decimation is appropriate. The filter threw away the out of band signal that would get aliased on top of your signal if you didn't filter. -- These are my opinions. I hate spam.

TV

Tom Van Baak

Thu, Jan 10, 2019 9:31 PM

I don't think case A makes sense. You are throwing away information. You
will get aliasing.

Hal,

Ah, but we do this all the time. Your GPSDO or your cesium standard outputs 10 MHz. There are many cases where throwing 9,999,999 of every 10,000,000 is useful. The result is ... 1PPS.

Yes, this is throwing away information. But most of it is useless information. For example, if your Cs is ahead of GPS by 123 ns, you don't need 10 million measurements to tell you that. One will do. So 1PPS is useful here.

And about aliasing, right, 1PPS is not immune to that. If your Cs clock secretly speed up by 5x for 200 milliseconds you would not see this crime in your 1PPS data. If you're worried about that happening, one solution is simply to measure at 10 PPS instead.

/tvb

> I don't think case A makes sense. You are throwing away information. You > will get aliasing. Hal, Ah, but we do this all the time. Your GPSDO or your cesium standard outputs 10 MHz. There are many cases where throwing 9,999,999 of every 10,000,000 is useful. The result is ... 1PPS. Yes, this is throwing away information. But most of it is useless information. For example, if your Cs is ahead of GPS by 123 ns, you don't need 10 million measurements to tell you that. One will do. So 1PPS is useful here. And about aliasing, right, 1PPS is not immune to that. If your Cs clock secretly speed up by 5x for 200 milliseconds you would not see this crime in your 1PPS data. If you're worried about that happening, one solution is simply to measure at 10 PPS instead. /tvb

MD

Magnus Danielson

Thu, Jan 10, 2019 9:41 PM

Hi,

On 2019-01-10 22:31, Tom Van Baak wrote:

I don't think case A makes sense. You are throwing away information. You
will get aliasing.

Hal,

Ah, but we do this all the time. Your GPSDO or your cesium standard outputs 10 MHz. There are many cases where throwing 9,999,999 of every 10,000,000 is useful. The result is ... 1PPS.

Yes, this is throwing away information. But most of it is useless information. For example, if your Cs is ahead of GPS by 123 ns, you don't need 10 million measurements to tell you that. One will do. So 1PPS is useful here.

Sure, but decimation methods may be applicable to estimate a good value
for it. It is not always needed, for sure, but this is one of the cases
you want to do one of them and others the others.

Cheers,
Magnus

Hi, On 2019-01-10 22:31, Tom Van Baak wrote: >> I don't think case A makes sense. You are throwing away information. You >> will get aliasing. > Hal, > > Ah, but we do this all the time. Your GPSDO or your cesium standard outputs 10 MHz. There are many cases where throwing 9,999,999 of every 10,000,000 is useful. The result is ... 1PPS. > > Yes, this is throwing away information. But most of it is useless information. For example, if your Cs is ahead of GPS by 123 ns, you don't need 10 million measurements to tell you that. One will do. So 1PPS is useful here. Sure, but decimation methods may be applicable to estimate a good value for it. It is not always needed, for sure, but this is one of the cases you want to do one of them and others the others. Cheers, Magnus

PK

Poul-Henning Kamp

Fri, Jan 11, 2019 8:24 AM

In message 293ce126-e9ef-40a2-e742-966d638cfda3@rubidium.se, Magnus Danielson
writes:

I can add that as of this morning, "decimate" is also used for the act
of reading only every Nth email in a long thread :-)

--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

-------- In message <293ce126-e9ef-40a2-e742-966d638cfda3@rubidium.se>, Magnus Danielson writes: I can add that as of this morning, "decimate" is also used for the act of reading only every Nth email in a long thread :-) -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 phk@FreeBSD.ORG | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence.

time-nuts@lists.febo.com

Re: [time-nuts] Misuse of word "decimate" (was Re: Short term 10MHz source)