Effect of OverSampling

Started by pratik13 June 21, 2006
I'm new to DSP and also this group.
My question is "what can be the negative effect of Oversamling in terms of
Destortion in original signal?". As in undersampling two spectrum are
overlapped and we may not be able to get original signal.. is that any
kind of effect in oversampling?


pratik13 wrote:
> I'm new to DSP and also this group. > My question is "what can be the negative effect of Oversamling in terms
of
> Destortion in original signal?". As in undersampling two spectrum are > overlapped and we may not be able to get original signal.. is that any > kind of effect in oversampling?
I don't know in theory if there is, but oversampling requires the consumption of extra resources, faster a/d, faster processor, more power, more time, more money, also the performance of a given a/d generally degrades at higher sampling rates (becomes more non-linear), and requires more talent to make the board work (layout becomes more critical), together these may cancel out any performance increase you were expecting.
pratik13 wrote:
> I'm new to DSP and also this group. > My question is "what can be the negative effect of Oversamling in terms
of
> Destortion in original signal?". As in undersampling two spectrum are > overlapped and we may not be able to get original signal.. is that any > kind of effect in oversampling?
The only drawback is the need for extra computation. Up to a point, it can make some computations easier. The original signal is what it is. If it contains distortion, that will be reproduced. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
pratik13 wrote:
> I'm new to DSP and also this group. > My question is "what can be the negative effect of Oversamling in terms
of
> Destortion in original signal?". As in undersampling two spectrum are > overlapped and we may not be able to get original signal.. is that any > kind of effect in oversampling? > >
As you probably learned early in your DSP work, you can only sample data with the highest frequency equal to half of your sampling rate. In order to achieve/guarantee this you typically need an anti-aliasing filter (i.e. an analog filter in front of the A/D). By oversampling the signal you can have a simpler, cheaper anti-aliasing filter and then implement a "brick wall filter" digitally which is cheap these days. On a related note this technique is built into sigma-delta converters. They do 1-bit sampling at a very fast rate, typically referred to as the modulation frequency. That data is then put through a digital filter to get the "normal" 16- or 24-bit data at the "sampling frequency" you're accustomed to seeing. Generally that digital filter will get rid of any frequencies above half the modulation frequency leaving you with a alias-free output. So in the case of sigma-delta converters you generally don't need an anti-aliasing filter as long as you don't have signal energy above half the modulation frequency. Brad
Brad Griffis wrote:

> On a related note this technique is built into sigma-delta converters. > They do 1-bit sampling at a very fast rate, typically referred to as the > modulation frequency. That data is then put through a digital filter to > get the "normal" 16- or 24-bit data at the "sampling
frequency" you're
> accustomed to seeing. Generally that digital filter will get rid of any > frequencies above half the modulation frequency leaving you with a > alias-free output. So in the case of sigma-delta converters you > generally don't need an anti-aliasing filter as long as you don't have > signal energy above half the modulation frequency. > > Brad
So using a low speed sigma-delta converter (say 1Khz conversion rate with 64Khz modulation frequency) would be similar in performance to using a higher speed SAR converter (say 64Khz) with oversampling/brick wall filter (assuming equivalent anti-aliasing filter)? I have read various technical notes concerning sigma-delta converters and can't really figure them out so I am unable to answer that question for myself. :( steve
steve wrote:
> Brad Griffis wrote: > >> On a related note this technique is built into sigma-delta converters. >> They do 1-bit sampling at a very fast rate, typically referred to as the >> modulation frequency. That data is then put through a digital filter to >> get the "normal" 16- or 24-bit data at the "sampling
frequency" you're
>> accustomed to seeing. Generally that digital filter will get rid of any >> frequencies above half the modulation frequency leaving you with a >> alias-free output. So in the case of sigma-delta converters you >> generally don't need an anti-aliasing filter as long as you don't have >> signal energy above half the modulation frequency. >> >> Brad > > So using a low speed sigma-delta converter (say 1Khz conversion rate > with 64Khz modulation frequency) would be similar in performance to > using a higher speed SAR converter (say 64Khz) with oversampling/brick > wall filter (assuming equivalent anti-aliasing filter)? > > I have read various technical notes concerning sigma-delta converters > and can't really figure them out so I am unable to answer that question > for myself. :( > > steve >
The "performance" involves a lot of things, i.e. you'd look at the linearity through INL/DNL in addition to things like ENoB. From a signal processing perspective you've got the right idea. A sigma delta converter usually has graphs showing what the frequency response of the digital filter looks like. In SD converters for applications like audio that will generally look close to a brick wall filter with the cutoff at half the modulation frequency. Keep in mind that you're only getting 1-bit data at the modulation frequency so it takes a lot of oversampling for the filtered data to turn into say 16-bits. Also, because you have a digital filter there is some settling time involved. For applications where you have a bunch of channels muxed into one converter you're generally better off with a SAR converter because you have to wait the specified settling time each time you change inputs to the converter. Brad
Brad Griffis wrote:

> ... For applications where you have a bunch of > channels muxed into one converter you're generally better off with a SAR > converter because you have to wait the specified settling time each time > you change inputs to the converter.
Also for servos and such where delay leads to instability. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Another possible negative side effect of  oversampling not mention so
far is regarding spectral resolution.

If you are dealing with finite storage (ie, memory), you will get
better resolution by sampling at a lower frequency but for a longer
period of time, as opposed to sampling at the highest frequency avaible
for the shortest period of time. Since the spectral resolution is a
function of both the number of samples and the sampling frequency,
sampling at a lower frequency for a given number of points will yield
better spectral resolution.


-Ikaro


pratik13 wrote:
> I'm new to DSP and also this group. > My question is "what can be the negative effect of Oversamling in terms
of
> Destortion in original signal?". As in undersampling two spectrum are > overlapped and we may not be able to get original signal.. is that any > kind of effect in oversampling?
I am afraid this information may be misleading for the novice. It is still
worth emphasizing that as long as the fixed variable is duration (instead
of the number of signal samples), hence number of samples increases
proportional to oversampling, there will be no loss of spectral
resolution.

>Another possible negative side effect of oversampling not mention so >far is regarding spectral resolution. > >If you are dealing with finite storage (ie, memory), you will get >better resolution by sampling at a lower frequency but for a longer >period of time, as opposed to sampling at the highest frequency avaible >for the shortest period of time. Since the spectral resolution is a >function of both the number of samples and the sampling frequency, >sampling at a lower frequency for a given number of points will yield >better spectral resolution. > > >-Ikaro
I am afraid this information may be misleading for the novice. It is still
worth emphasizing that as long as the fixed variable is duration (instead
of the number of signal samples), hence number of samples increases
proportional to oversampling, there will be no loss of spectral
resolution.

>Another possible negative side effect of oversampling not mention so >far is regarding spectral resolution. > >If you are dealing with finite storage (ie, memory), you will get >better resolution by sampling at a lower frequency but for a longer >period of time, as opposed to sampling at the highest frequency avaible >for the shortest period of time. Since the spectral resolution is a >function of both the number of samples and the sampling frequency, >sampling at a lower frequency for a given number of points will yield >better spectral resolution. > > >-Ikaro
Another possible negative side effect of  oversampling not mention so
far is regarding spectral resolution.

If you are dealing with finite storage (ie, memory), you will get
better resolution by sampling at a lower frequency but for a longer
period of time, as opposed to sampling at the highest frequency avaible
for the shortest period of time. Since the spectral resolution is a
function of both the number of samples and the sampling frequency,
sampling at a lower frequency for a given number of points will yield
better spectral resolution.


-Ikaro


pratik13 wrote:
> I'm new to DSP and also this group. > My question is "what can be the negative effect of Oversamling in terms
of
> Destortion in original signal?". As in undersampling two spectrum are > overlapped and we may not be able to get original signal.. is that any > kind of effect in oversampling?
Brad Griffis wrote:

> ... For applications where you have a bunch of > channels muxed into one converter you're generally better off with a SAR > converter because you have to wait the specified settling time each time > you change inputs to the converter.
Also for servos and such where delay leads to instability. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
steve wrote:
> Brad Griffis wrote: > >> On a related note this technique is built into sigma-delta converters. >> They do 1-bit sampling at a very fast rate, typically referred to as the >> modulation frequency. That data is then put through a digital filter to >> get the "normal" 16- or 24-bit data at the "sampling
frequency" you're
>> accustomed to seeing. Generally that digital filter will get rid of any >> frequencies above half the modulation frequency leaving you with a >> alias-free output. So in the case of sigma-delta converters you >> generally don't need an anti-aliasing filter as long as you don't have >> signal energy above half the modulation frequency. >> >> Brad > > So using a low speed sigma-delta converter (say 1Khz conversion rate > with 64Khz modulation frequency) would be similar in performance to > using a higher speed SAR converter (say 64Khz) with oversampling/brick > wall filter (assuming equivalent anti-aliasing filter)? > > I have read various technical notes concerning sigma-delta converters > and can't really figure them out so I am unable to answer that question > for myself. :( > > steve >
The "performance" involves a lot of things, i.e. you'd look at the linearity through INL/DNL in addition to things like ENoB. From a signal processing perspective you've got the right idea. A sigma delta converter usually has graphs showing what the frequency response of the digital filter looks like. In SD converters for applications like audio that will generally look close to a brick wall filter with the cutoff at half the modulation frequency. Keep in mind that you're only getting 1-bit data at the modulation frequency so it takes a lot of oversampling for the filtered data to turn into say 16-bits. Also, because you have a digital filter there is some settling time involved. For applications where you have a bunch of channels muxed into one converter you're generally better off with a SAR converter because you have to wait the specified settling time each time you change inputs to the converter. Brad
Brad Griffis wrote:

> On a related note this technique is built into sigma-delta converters. > They do 1-bit sampling at a very fast rate, typically referred to as the > modulation frequency. That data is then put through a digital filter to > get the "normal" 16- or 24-bit data at the "sampling
frequency" you're
> accustomed to seeing. Generally that digital filter will get rid of any > frequencies above half the modulation frequency leaving you with a > alias-free output. So in the case of sigma-delta converters you > generally don't need an anti-aliasing filter as long as you don't have > signal energy above half the modulation frequency. > > Brad
So using a low speed sigma-delta converter (say 1Khz conversion rate with 64Khz modulation frequency) would be similar in performance to using a higher speed SAR converter (say 64Khz) with oversampling/brick wall filter (assuming equivalent anti-aliasing filter)? I have read various technical notes concerning sigma-delta converters and can't really figure them out so I am unable to answer that question for myself. :( steve
pratik13 wrote:
> I'm new to DSP and also this group. > My question is "what can be the negative effect of Oversamling in terms
of
> Destortion in original signal?". As in undersampling two spectrum are > overlapped and we may not be able to get original signal.. is that any > kind of effect in oversampling? > >
As you probably learned early in your DSP work, you can only sample data with the highest frequency equal to half of your sampling rate. In order to achieve/guarantee this you typically need an anti-aliasing filter (i.e. an analog filter in front of the A/D). By oversampling the signal you can have a simpler, cheaper anti-aliasing filter and then implement a "brick wall filter" digitally which is cheap these days. On a related note this technique is built into sigma-delta converters. They do 1-bit sampling at a very fast rate, typically referred to as the modulation frequency. That data is then put through a digital filter to get the "normal" 16- or 24-bit data at the "sampling frequency" you're accustomed to seeing. Generally that digital filter will get rid of any frequencies above half the modulation frequency leaving you with a alias-free output. So in the case of sigma-delta converters you generally don't need an anti-aliasing filter as long as you don't have signal energy above half the modulation frequency. Brad
pratik13 wrote:
> I'm new to DSP and also this group. > My question is "what can be the negative effect of Oversamling in terms
of
> Destortion in original signal?". As in undersampling two spectrum are > overlapped and we may not be able to get original signal.. is that any > kind of effect in oversampling?
The only drawback is the need for extra computation. Up to a point, it can make some computations easier. The original signal is what it is. If it contains distortion, that will be reproduced. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
pratik13 wrote:
> I'm new to DSP and also this group. > My question is "what can be the negative effect of Oversamling in terms
of
> Destortion in original signal?". As in undersampling two spectrum are > overlapped and we may not be able to get original signal.. is that any > kind of effect in oversampling?
I don't know in theory if there is, but oversampling requires the consumption of extra resources, faster a/d, faster processor, more power, more time, more money, also the performance of a given a/d generally degrades at higher sampling rates (becomes more non-linear), and requires more talent to make the board work (layout becomes more critical), together these may cancel out any performance increase you were expecting.
I'm new to DSP and also this group.
My question is "what can be the negative effect of Oversamling in terms of
Destortion in original signal?". As in undersampling two spectrum are
overlapped and we may not be able to get original signal.. is that any
kind of effect in oversampling?