FPGA Central

Hi all,

I need to sample a 100 Hz - 15 kHz analog FM audio (music and speech)
channel for an unusual purpose - to measure the time that certain audio
events occur.

Audio quality, resolution, distortion are only important to the extent they
affect the time accuracy of event detections.

I'd like to be able to detect the occurrence of specific audio events in the
channel with an accuracy of better than one microsecond. This will be a new
design so there are no restrictions on sampling rate, choice of ADC etc.

Can anyone point me in the right direction please?

Thanks in advance.

"Steamer" <[email protected]> wrote in message
news:UOTli.21839$[email protected] et...
> Hi all,
>
> I need to sample a 100 Hz - 15 kHz analog FM audio (music and speech)
> channel for an unusual purpose - to measure the time that certain audio
> events occur.
>
> Audio quality, resolution, distortion are only important to the extent
they
> affect the time accuracy of event detections.
>
> I'd like to be able to detect the occurrence of specific audio events in
the
> channel with an accuracy of better than one microsecond. This will be a
new
> design so there are no restrictions on sampling rate, choice of ADC etc.

Measuring arrival time with the accuracy of 1/100 of the bandwidth is not
going to be very simple. You may need specially crafted signals and a
precisely calibrated channel.

> Can anyone point me in the right direction please?

What exactly are you trying to do?

Vladimir Vassilevsky
www.abvolt.com

"Vladimir Vassilevsky" <[email protected]> wrote in message
news:itXli.11302$[email protected] et...

>snip<

> What exactly are you trying to do?
>

What I'm trying to do is to measure the amount of time shift between two
audio channels that are identical but shifted in time from one another
(because the delayed signal has passed through equipment whose propagation
time we wish to determine).

There is a detector on each of the two audio paths. In each detector,
analog audio feeds the ADC, then a processor looks at the digitized audio
for a pre-specified signal event that will be used as a time marker. When
that event occurs, the processor recognizes it and generates a strobe
output. The time difference in strobe outputs of the two procesors
represents the time shift between the two audio paths.

The amount of time needed to detect the signal event and generate the strobe
doesn't matter, as long as it's the same - to better than a microseond - in
both detectors. We expect something around a microsecond of audio delay, so
we need a measurement accuracy better than that.

Appreciate your help with this problem.

Posted from another office computer, sorry.

"George" <[email protected]> wrote in message
news:fLYli.46286$[email protected] net...
>
> "Vladimir Vassilevsky" <[email protected]> wrote in message
> news:itXli.11302$[email protected] et...
>
>>snip<
>
>> What exactly are you trying to do?
>>
>
> What I'm trying to do is to measure the amount of time shift between two
> audio channels that are identical but shifted in time from one another
> (because the delayed signal has passed through equipment whose propagation
> time we wish to determine).
>
> There is a detector on each of the two audio paths. In each detector,
> analog audio feeds the ADC, then a processor looks at the digitized audio
> for a pre-specified signal event that will be used as a time marker. When
> that event occurs, the processor recognizes it and generates a strobe
> output. The time difference in strobe outputs of the two procesors
> represents the time shift between the two audio paths.
>
> The amount of time needed to detect the signal event and generate the
> strobe doesn't matter, as long as it's the same - to better than a
> microseond - in both detectors. We expect something around a microsecond
> of audio delay, so we need a measurement accuracy better than that.
>
> Appreciate your help with this problem.
>
>

George schrieb:

> What I'm trying to do is to measure the amount of time shift between two
> audio channels that are identical but shifted in time from one another
> (because the delayed signal has passed through equipment whose propagation
> time we wish to determine).
>
> There is a detector on each of the two audio paths. In each detector,
> analog audio feeds the ADC, then a processor looks at the digitized audio
> for a pre-specified signal event that will be used as a time marker. When
> that event occurs, the processor recognizes it and generates a strobe
> output. The time difference in strobe outputs of the two procesors
> represents the time shift between the two audio paths.

It should be easier and more accurate to sample both signals for some time
using the same clock and calculate a cross correlation between them to get
the delay. You may sample both signals using a stereo ADC (sound card?)
to have a common time base.


Hendrik vdH

George wrote:
> "Vladimir Vassilevsky" <[email protected]> wrote in message
> news:itXli.11302$[email protected] et...
>
>> snip<
>
>> What exactly are you trying to do?
>>
>
> What I'm trying to do is to measure the amount of time shift between two
> audio channels that are identical but shifted in time from one another
> (because the delayed signal has passed through equipment whose propagation
> time we wish to determine).
>
> There is a detector on each of the two audio paths. In each detector,
> analog audio feeds the ADC, then a processor looks at the digitized audio
> for a pre-specified signal event that will be used as a time marker. When
> that event occurs, the processor recognizes it and generates a strobe
> output. The time difference in strobe outputs of the two procesors
> represents the time shift between the two audio paths.
>
> The amount of time needed to detect the signal event and generate the strobe
> doesn't matter, as long as it's the same - to better than a microseond - in
> both detectors. We expect something around a microsecond of audio delay, so
> we need a measurement accuracy better than that.
>
> Appreciate your help with this problem.

First, how does FM come in? If your audio baseband, of modulated in some
way? If modulated, what it the carrier?

Assuming baseband, the technique you want is cross correlation. No
special event markers are needed. In fact, broadband noise is an
excellent test signal for you.

Jerry
--
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

"Jerry Avins" <[email protected]> wrote in message
news:[email protected]..

> > What I'm trying to do is to measure the amount of time shift between two
> > audio channels that are identical but shifted in time from one another
> > (because the delayed signal has passed through equipment whose
propagation
> > time we wish to determine).
> >
>
> Assuming baseband, the technique you want is cross correlation. No
> special event markers are needed. In fact, broadband noise is an
> excellent test signal for you.

Cross correlation of the noise like signals would be very demanding in the
computation and not optimal for the SNR. I would measure the phase shift at
the sinewave of the highest frequency. Resolving the ambiguity of +/- number
of periods is not a problem either.

Vladimir Vassilevsky
www.abvolt.com

Thanks to those above who have offered comments, but I see my explanation
may have been misleading.

The two audio signals, identical but with one delayed in time, are not
available for measurement at the same physical location, which is what makes
the problem interesting. They are at different locations. We want to
measure audio path propagation time through two sets of equipment on two
different routes at two different destinations. A common highly-accurate
time reference is available in both locations.

Our goal is to detect a unique event in the audio path and to record its
time of occurrence relative to our common time reference, at each
destination. The difference in detection times represents the relative
delay in the two audio paths.

So I think the key question is: how accurately can a pre-defined audio event
(a data pattern after digitizing the audio) be time-tagged? We are not
limited by noise, ADC resolution or sampling rate.

Thank you.




"George" <[email protected]> wrote in message
news:fLYli.46286$[email protected] net...

> What I'm trying to do is to measure the amount of time shift between two
> audio channels that are identical but shifted in time from one another
> (because the delayed signal has passed through equipment whose propagation
> time we wish to determine).
>
> There is a detector on each of the two audio paths. In each detector,
> analog audio feeds the ADC, then a processor looks at the digitized audio
> for a pre-specified signal event that will be used as a time marker. When
> that event occurs, the processor recognizes it and generates a strobe
> output. The time difference in strobe outputs of the two procesors
> represents the time shift between the two audio paths.
>
> The amount of time needed to detect the signal event and generate the
> strobe doesn't matter, as long as it's the same - to better than a
> microseond - in both detectors. We expect something around a microsecond
> of audio delay, so we need a measurement accuracy better than that.
>
> Appreciate your help with this problem.
>
>

"Steamer" <[email protected]> wrote in message
news:Ji9mi.766$[email protected] ..

> The two audio signals, identical but with one delayed in time, are not
> available for measurement at the same physical location, which is what
makes
> the problem interesting. They are at different locations. We want to
> measure audio path propagation time through two sets of equipment on two
> different routes at two different destinations. A common highly-accurate
> time reference is available in both locations.
>
> Our goal is to detect a unique event in the audio path and to record its
> time of occurrence relative to our common time reference, at each
> destination. The difference in detection times represents the relative
> delay in the two audio paths.
>
> So I think the key question is: how accurately can a pre-defined audio
event
> (a data pattern after digitizing the audio) be time-tagged? We are not
> limited by noise, ADC resolution or sampling rate.

If there is no noise and no distortion, then you can measure the time of
arrival with the infinite accuracy. So, it comes to the question of how good
is the actual channel. It also depends on the pattern that you are
triggering on. Taking the reasonable assumptions, it looks fairly simple to
achieve the accuracy of tens of microseconds, however it would not be very
easy to get to 1uS.

Vladimir Vassilevsky
DSP and Mixed Signal Consultant
www.abvolt.com

In article <W6qmi.39506$[email protected]> , "Vladimir Vassilevsky" <[email protected]> wrote:
>
>"Steamer" <[email protected]> wrote in message
>news:Ji9mi.766$[email protected] ...
>
>> The two audio signals, identical but with one delayed in time, are not
>> available for measurement at the same physical location, which is what
>makes
>> the problem interesting. They are at different locations. We want to
>> measure audio path propagation time through two sets of equipment on two
>> different routes at two different destinations. A common highly-accurate
>> time reference is available in both locations.
>>
>> Our goal is to detect a unique event in the audio path and to record its
>> time of occurrence relative to our common time reference, at each
>> destination. The difference in detection times represents the relative
>> delay in the two audio paths.
>>
>> So I think the key question is: how accurately can a pre-defined audio
>event
>> (a data pattern after digitizing the audio) be time-tagged? We are not
>> limited by noise, ADC resolution or sampling rate.
>
>If there is no noise and no distortion, then you can measure the time of
>arrival with the infinite accuracy. So, it comes to the question of how good
>is the actual channel. It also depends on the pattern that you are
>triggering on. Taking the reasonable assumptions, it looks fairly simple to
>achieve the accuracy of tens of microseconds, however it would not be very
>easy to get to 1uS.
>
>Vladimir Vassilevsky
>DSP and Mixed Signal Consultant
>www.abvolt.com
>
And how accurately do you know the time? I had a lecture on time long, long
ago and it's not that easy to get time as seperate locations to 250
nanoseconds of accuracy.

And as implied above, what is the signal to noise ratio? It would have to be
very high, maybe 60 dB, maybe more.