How to compress large signal for display

I have 100.000 samples long signal I need to present on the display (MS 
Windows using DirectX).

Simple 'use every n-th sample' or downsampling method is not good enough. 
How to compress signal so it can be presented within few hundred pixels on 
the screen? If possible algorithm should be quite fast as I need to update 
several traces (tens) and display them few times per second.

Thanks,

Damir

in article dm00po$cvq$1@ss405.t-com.hr, damir at dzagar@BRISIsrce.hr wrote
on 11/22/2005 15:58:

> I have 100.000 samples long signal I need to present on the display (MS
> Windows using DirectX).

for those of us that are North American bound, the number above is 10^5, not
10^2 .  for some reason, we use commas where everyone else uses periods to
delimite 1000s.

> Simple 'use every n-th sample' or downsampling method is not good enough.

correct.  it isn't.

> How to compress signal so it can be presented within few hundred pixels on
> the screen?

this is the way most audio editing software does it: for all audio samples
that map to a particular pixel, the minimum sample value is determined and
the maximum value is determined, and a vertical line is drawn from the
minimum value to the maximum value at that pixel location.  that way, as you
zoom in, the apparent envelope (that might not be symmetric if the waveform
is not) actually begins to look like the waveform and does, in the limit,
become the waveform when you get to a scale of 1 pixel per sample.  if you
zoom in even more (more than 1 pixel per sample), if you want the waveform
to look nice, you have to interpolate using either some polynomial
interpolation (like Hermite or Lagrange) or a windowed sinc().

> If possible algorithm should be quite fast as I need to update
> several traces (tens) and display them few times per second.

it's probably not way too fast, but, if you're doing this real-time, you can
always maintain a running maximum value and running minimum value.

-- 

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

"damir" <dzagar@BRISIsrce.hr> wrote in message
news:dm00po$cvq$1@ss405.t-com.hr...
> I have 100.000 samples long signal I need to present on the display (MS
> Windows using DirectX).
>
> Simple 'use every n-th sample' or downsampling method is not good enough.

Who not do an average of every n samples?

> How to compress signal so it can be presented within few hundred pixels on
> the screen? If possible algorithm should be quite fast as I need to update
> several traces (tens) and display them few times per second.
>
> Thanks,
>
> Damir
>
>

damir <dzagar@brisisrce.hr> wrote:
> I have 100.000 samples long signal I need to present on the display (MS 
> Windows using DirectX).
> 
> Simple 'use every n-th sample' or downsampling method is not good enough. 
> How to compress signal so it can be presented within few hundred pixels on 
> the screen? If possible algorithm should be quite fast as I need to update 
> several traces (tens) and display them few times per second.


upper and lower envelopes (max & min from every n samples)

damir wrote:
> I have 100.000 samples long signal I need to present on the display (MS 
> Windows using DirectX).
> 
> Simple 'use every n-th sample' or downsampling method is not good enough. 
> How to compress signal so it can be presented within few hundred pixels on 
> the screen? If possible algorithm should be quite fast as I need to update 
> several traces (tens) and display them few times per second.

Compressing 100,000 samples to a few hundred is downsampling by 
definition I hope it is just that downsampling /the way you went about 
it/ had problems. You're out of luck if there is no better way.

Maybe we can help if you tell us what you tried and why you don't like 
the result.

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

in article 43838bf4$0$41137$14726298@news.sunsite.dk, Bhaskar Thiagarajan at
bhaskart@deja.com wrote on 11/22/2005 16:21:

> "damir" <dzagar@BRISIsrce.hr> wrote in message
> news:dm00po$cvq$1@ss405.t-com.hr...
>> I have 100.000 samples long signal I need to present on the display (MS
>> Windows using DirectX).
>> 
>> Simple 'use every n-th sample' or downsampling method is not good enough.
> 
> Wh[y] not do an average of every n samples?
> 

so a sine wave gets displayed as DC or zero?

-- 

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

"robert bristow-johnson" <rbj@audioimagination.com> wrote in message
news:BFA90CE6.C5D2%rbj@audioimagination.com...
> in article 43838bf4$0$41137$14726298@news.sunsite.dk, Bhaskar Thiagarajan
at
> bhaskart@deja.com wrote on 11/22/2005 16:21:
>
> > "damir" <dzagar@BRISIsrce.hr> wrote in message
> > news:dm00po$cvq$1@ss405.t-com.hr...
> >> I have 100.000 samples long signal I need to present on the display (MS
> >> Windows using DirectX).
> >>
> >> Simple 'use every n-th sample' or downsampling method is not good
enough.
> >
> > Wh[y] not do an average of every n samples?
> >
>
> so a sine wave gets displayed as DC or zero?

The OP didn't say what his signal content was. I was going to respond with a
whole bunch of choices including the method you suggested but reverted to
keeping it simple.

This is a common problem in several applications including displaying data
in spectrum analyzers, where, the term 'detection' is used for the method of
how data gets mapped into display pixels. And in that situation, when
'average' is used, CW or sinusoids suffer from rather inaccurate amplitude
information. The choice of 'detection' is left to the user depending on the
type of signal he wants to observe (in some cases, modern spectrum analyzers
attempt to do some level of automation based on the type of measurement the
user is performing).

So the question to the OP is, what kind of signals need to be presented on
the display and why does 'use every n-th sample' not work for you?

Cheers
Bhaskar

>
> -- 
>
> r b-j                  rbj@audioimagination.com
>
> "Imagination is more important than knowledge."
>
>

> for those of us that are North American bound, the number above is 10^5, 
> not
> 10^2 .  for some reason, we use commas where everyone else uses periods to
> delimite 1000s.

For sure, my intention was not to write 100 using 3 decimal digits :))

> this is the way most audio editing software does it: for all audio samples
> that map to a particular pixel, the minimum sample value is determined and
> the maximum value is determined, and a vertical line is drawn from the
> minimum value to the maximum value at that pixel location.  that way, as 
> you
> zoom in, the apparent envelope (that might not be symmetric if the 
> waveform
> is not) actually begins to look like the waveform and does, in the limit,
> become the waveform when you get to a scale of 1 pixel per sample.  if you
> zoom in even more (more than 1 pixel per sample), if you want the waveform
> to look nice, you have to interpolate using either some polynomial
> interpolation (like Hermite or Lagrange) or a windowed sinc().
>

I thought about something like that - calculate envelope for min. and max. 
values. I will look for some algorithm that may be used for "real-time" 
application.

Thanks,

Damir

Jerry Avins wrote:
> damir wrote:
> > I have 100.000 samples long signal I need to present on the display (MS
> > Windows using DirectX).
> >
> > Simple 'use every n-th sample' or downsampling method is not good enough.
> > How to compress signal so it can be presented within few hundred pixels on
> > the screen? If possible algorithm should be quite fast as I need to update
> > several traces (tens) and display them few times per second.
>
> Compressing 100,000 samples to a few hundred is downsampling by
> definition I hope it is just that downsampling /the way you went about
> it/ had problems. You're out of luck if there is no better way.

For visual purposes, one probably wants to downsample in
two dimensions, not just one.  Maybe grayscale representing
the number of samples in each volts-seconds plot rectangle,
perhaps 2d lowpass filtered.


IMHO. YMMV.
-- 
rhn A.T nicholson d.O.t C-o-M

damir wrote:

> I have 100.000 samples long signal I need to present on the display (MS 
> Windows using DirectX).

> Simple 'use every n-th sample' or downsampling method is not good enough. 
> How to compress signal so it can be presented within few hundred pixels on 
> the screen? If possible algorithm should be quite fast as I need to update 
> several traces (tens) and display them few times per second.

The usual methods are based on estimates of the second derivative.

I first saw this problem in taking data to be sent to a plotter based on 
stepper motors, where each step was written to a tape, and displaying it 
on a vector CRT display with limited memory.

One I did once, to be as simple as possible to write, was to calculate 
over the set of points the absolute value of the second difference. 
Then find the points, possibly interpolated, that divide the area under 
the curve into equal intervals.  It is a few DO loops in Fortran over 
the arrays.

-- glen