Time/Frequency

There is an input of strong pulse closely followed by useful small 
signal. Strong pulse initial amplitude is ~1e5 times that of signal; 
however it falls exponentially to essentially zero in duration of ~5 
samples of ADC.

I need to measure the signal; ignoring preceding strong pulse.
Time point where the strong pulse ends and the useful signal starts is 
known.

Problem: input is sampled by delta-sigma ADC. Therefore strong pulse is 
smeared over ~16 samples by delta-sigma internal lowpass FIR filter, so 
it gets superposed with useful signal.

Is there a way to separate signals after ADC ?

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

On Fri, 13 Dec 2013 11:20:57 -0600, Vladimir Vassilevsky wrote:

> There is an input of strong pulse closely followed by useful small
> signal. Strong pulse initial amplitude is ~1e5 times that of signal;
> however it falls exponentially to essentially zero in duration of ~5
> samples of ADC.
> 
> I need to measure the signal; ignoring preceding strong pulse.
> Time point where the strong pulse ends and the useful signal starts is
> known.
> 
> Problem: input is sampled by delta-sigma ADC. Therefore strong pulse is
> smeared over ~16 samples by delta-sigma internal lowpass FIR filter, so
> it gets superposed with useful signal.
> 
> Is there a way to separate signals after ADC ?

You didn't say how close in time is "closely".

The second thing that I'd consider trying would be to deconvolve the 
delta-sigma's internal FIR, at least partially, to get prompter response.

The first thing that I'd consider trying would be to change something in 
the physical measurement setup -- i.e. the ADC, a blanker in front of the 
ADC, change to a delta-sigma with a different FIR filter or a 
programmable one, etc.

I assume that the first thing has already occurred to you, and you have 
good reasons for not being able to do it.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

On 12/13/2013 11:29 AM, Tim Wescott wrote:
> On Fri, 13 Dec 2013 11:20:57 -0600, Vladimir Vassilevsky wrote:
>
>> There is an input of strong pulse closely followed by useful small
>> signal. Strong pulse initial amplitude is ~1e5 times that of signal;
>> however it falls exponentially to essentially zero in duration of ~5
>> samples of ADC.
>>
>> I need to measure the signal; ignoring preceding strong pulse.
>> Time point where the strong pulse ends and the useful signal starts is
>> known.
>>
>> Problem: input is sampled by delta-sigma ADC. Therefore strong pulse is
>> smeared over ~16 samples by delta-sigma internal lowpass FIR filter, so
>> it gets superposed with useful signal.
>>
>> Is there a way to separate signals after ADC ?
>
> You didn't say how close in time is "closely".

Read up:

 >> however it falls exponentially to essentially zero in duration of ~5
 >> samples of ADC.

> The second thing that I'd consider trying would be to deconvolve the
> delta-sigma's internal FIR, at least partially, to get prompter response.

Undo brickwall FIR filter. Especially after decimation.
Great idea.

> The first thing that I'd consider trying would be to change something in
> the physical measurement setup -- i.e. the ADC, a blanker in front of the
> ADC, change to a delta-sigma with a different FIR filter or a
> programmable one, etc.

I.e. buy a new house. Great idea, too.

> I assume that the first thing has already occurred to you, and you have
> good reasons for not being able to do it.

It's been a while you wrote anything meaningful.

VLV

On 12/13/13 12:41 PM, Vladimir Vassilevsky wrote:
>
> It's been a while [Tim] wrote anything meaningful.
>

yeah, about 48 hours (12/11/13 3:43 PM, unless i missed something).



-- 

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

On 13.12.13 19:20, Vladimir Vassilevsky wrote:
> There is an input of strong pulse closely followed by useful small
> signal. Strong pulse initial amplitude is ~1e5 times that of signal;
> however it falls exponentially to essentially zero in duration of ~5
> samples of ADC.
>
> I need to measure the signal; ignoring preceding strong pulse.
> Time point where the strong pulse ends and the useful signal starts is
> known.
>
> Problem: input is sampled by delta-sigma ADC. Therefore strong pulse is
> smeared over ~16 samples by delta-sigma internal lowpass FIR filter, so
> it gets superposed with useful signal.
>
> Is there a way to separate signals after ADC ?
>
> Vladimir Vassilevsky
> DSP and Mixed Signal Designs
> www.abvolt.com
>

This is notably similar to the pulse (e.g. ignition) noise problem in
radio receivers. As soon as you have lost the frequency components 
outside the decimation filters, you need a guessing filter to get them back.

IMHO, you should electronically limit the input amplitude before the A/D 
conversion.

-- 

Tauno Voipio

On Fri, 13 Dec 2013 11:41:09 -0600, Vladimir Vassilevsky wrote:

> On 12/13/2013 11:29 AM, Tim Wescott wrote:
>> On Fri, 13 Dec 2013 11:20:57 -0600, Vladimir Vassilevsky wrote:
>>
>>> There is an input of strong pulse closely followed by useful small
>>> signal. Strong pulse initial amplitude is ~1e5 times that of signal;
>>> however it falls exponentially to essentially zero in duration of ~5
>>> samples of ADC.
>>>
>>> I need to measure the signal; ignoring preceding strong pulse.
>>> Time point where the strong pulse ends and the useful signal starts is
>>> known.
>>>
>>> Problem: input is sampled by delta-sigma ADC. Therefore strong pulse
>>> is smeared over ~16 samples by delta-sigma internal lowpass FIR
>>> filter, so it gets superposed with useful signal.
>>>
>>> Is there a way to separate signals after ADC ?
>>
>> You didn't say how close in time is "closely".
> 
> Read up:
> 
>  >> however it falls exponentially to essentially zero in duration of ~5
>  >> samples of ADC.
> 
>> The second thing that I'd consider trying would be to deconvolve the
>> delta-sigma's internal FIR, at least partially, to get prompter
>> response.
> 
> Undo brickwall FIR filter. Especially after decimation.
> Great idea.
> 
>> The first thing that I'd consider trying would be to change something
>> in the physical measurement setup -- i.e. the ADC, a blanker in front
>> of the ADC, change to a delta-sigma with a different FIR filter or a
>> programmable one, etc.
> 
> I.e. buy a new house. Great idea, too.
> 
>> I assume that the first thing has already occurred to you, and you have
>> good reasons for not being able to do it.
> 
> It's been a while you wrote anything meaningful.

http://en.wikipedia.org/wiki/Ad_hominem

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Is the strong pulse the same amplitude and shape every time? Is it repetitive ? If so, what is the interval between pulses; is it long enough that the filter history is flushed out before the next pulse comes along?

Bob

On 12/13/2013 4:05 PM, radams2000@gmail.com wrote:

> Is the strong pulse the same amplitude and shape every time?

Ideally, it should be. So it could be subtracted from result. 
Unfortunately, there are also random variations caused by temperature, 
supply voltage, etc. If I calibrate for the pulse once, it works great 
for some time until something shifts away. With the signal amplitude 
ratio of 1e5, compensation has to be very accurate.

> Is it
> repetitive ? If so, what is the interval between pulses; is it long
> enough that the filter history is flushed out before the next pulse
> comes along?

It is repetitive; history is flushed out completely at every run.

VLV

On 12/13/13 5:45 PM, Vladimir Vassilevsky wrote:
> On 12/13/2013 4:05 PM, radams2000@gmail.com wrote:
>
>> Is the strong pulse the same amplitude and shape every time?
>
> Ideally, it should be. So it could be subtracted from result.
> Unfortunately, there are also random variations caused by temperature,
> supply voltage, etc. If I calibrate for the pulse once, it works great
> for some time until something shifts away. With the signal amplitude
> ratio of 1e5, compensation has to be very accurate.
>
>> Is it
>> repetitive ? If so, what is the interval between pulses; is it long
>> enough that the filter history is flushed out before the next pulse
>> comes along?
>
> It is repetitive; history is flushed out completely at every run.

i wonder if a good pitch detector controlling a hard-core comb filter 
with nulls at harmonics of the pulse rate might do as well as anything 
to take out the pulse train?  delay line in the comb filter would need 
to be controllable to a fractional-sample precision.  and you could have 
multiple delay elements so you can design the shape of the teeth of the 
comb filter.

only guess i have.  please don't call me an idiot, Vlad.  could hurt my 
feelings.

but you can call Bob one.  i could use a chuckle.

:-)

-- 

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

On 13.12.13 18.20, Vladimir Vassilevsky wrote:
> I need to measure the signal; ignoring preceding strong pulse.
> Time point where the strong pulse ends and the useful signal starts is
> known.
>
> Problem: input is sampled by delta-sigma ADC. Therefore strong pulse is
> smeared over ~16 samples by delta-sigma internal lowpass FIR filter, so
> it gets superposed with useful signal.
>
> Is there a way to separate signals after ADC ?

No. Once the ADC is driven into non-linearity by exceeding the maximum 
slew rate of your ADC in your case the game is over.

Restrict the analog signal to values your ADC can capture, e.g. by 
introducing a 1st order low pass that blurs the signal to more samples. 
The result of this low pass then can be removed by a digital pole zero 
cancellation filter. This returns the pulse into the Dirac like shape 
that can be separated in the time domain.

Of course, this will introduce some extra noise into the high frequency 
components of your result. That's the price for the too slow delta sigma 
ADC.

If noise at high frequencies is not desirable you might alternatively 
use phase distortion with an analog filter like an all-pass. This may 
also widen your pulse in a reversible way.


Marcel