Forums

Should AWGN be part of a simulation setup that tests Adjacent Channel Interference performance?

Started by Andrew FPGA February 17, 2008
Hi there,
I'm trying to simulate the adjacent channel(ACI) performance of a
continuous phase modulation(CPM) scheme so that I can determine if my
modulator/demodulator meets a particular ETSI performance
specification. I'm not sure if my simulation should include AWGN as
well as the ACI? Ultimately I'm trying to chose a combination of CPM
modulation parameters (alphabet size, pulse shape, pulse length,
modulation index) that meet a specific power spectral density channel
mask, SNR performance, and adjacent channel interference performance.

My setup:
data source -> CPM modulator -> AWGN -> CPM demodulator -> error
counter

By changing the AWGN eb/no and measuring the BER I can generate BER vs
eb/no curves that match the literature.

For ACI, the ETSI specifcation says "For a carrier to interference
ratio of 0dB, a maximum degradation in the Receive input Signal Level
threshold(RSL) of 1 dB is allowed".

The RSL spec essentially says the receiver must achieve a minimum BER
of 10^-6 at an eb/no of 15dB.

In order to test the ACI performance, the first thing I did was create
ACI of the same power as the orignal signal(0dB) by duplicating the
data source and CPM modulator, and frequency shifting the result so it
lies in the adjacent channel. I then add the ACI to the original
signal after the AWGN in the above diagram.

How do I check that the RSL(eb/no) degradation is less than 1dB? Do I
reduce the AWGN by 1dB and check that the BER is less than or equal to
the BER without the ACI?

Should I even have AWGN in my ACI simulation? - I notice that as the
ACI gets stronger, the AWGN has less effect on BER, and at a certain
point the AWGN has no effect on BER - I suppose the ACI is
dominating.

Cheers
Andrew


On Feb 17, 4:31 am, Andrew FPGA <andrew.newsgr...@gmail.com> wrote:
> Hi there, > I'm trying to simulate the adjacent channel(ACI) performance of a > continuous phase modulation(CPM) scheme so that I can determine if my > modulator/demodulator meets a particular ETSI performance > specification. I'm not sure if my simulation should include AWGN as > well as the ACI? Ultimately I'm trying to chose a combination of CPM > modulation parameters (alphabet size, pulse shape, pulse length, > modulation index) that meet a specific power spectral density channel > mask, SNR performance, and adjacent channel interference performance. > > My setup: > data source -> CPM modulator -> AWGN -> CPM demodulator -> error > counter > > By changing the AWGN eb/no and measuring the BER I can generate BER vs > eb/no curves that match the literature. > > For ACI, the ETSI specifcation says "For a carrier to interference > ratio of 0dB, a maximum degradation in the Receive input Signal Level > threshold(RSL) of 1 dB is allowed". > > The RSL spec essentially says the receiver must achieve a minimum BER > of 10^-6 at an eb/no of 15dB. > > In order to test the ACI performance, the first thing I did was create > ACI of the same power as the orignal signal(0dB) by duplicating the > data source and CPM modulator, and frequency shifting the result so it > lies in the adjacent channel. I then add the ACI to the original > signal after the AWGN in the above diagram. > > How do I check that the RSL(eb/no) degradation is less than 1dB? Do I > reduce the AWGN by 1dB and check that the BER is less than or equal to > the BER without the ACI? > > Should I even have AWGN in my ACI simulation? - I notice that as the > ACI gets stronger, the AWGN has less effect on BER, and at a certain > point the AWGN has no effect on BER - I suppose the ACI is > dominating. > > Cheers > Andrew
For tests like this I would include AWGN, because it's there in the real world. First, I would turn off the ACI and adjust signal level for target BER (1-e6 in your case). Now turn on the ACI at equal power level to signal. The BER should get worse. Now turn the desired signal up until BER returns to 1e-6, leaving the ACI alone. If you turned it up < 1 dB, test passed. John
On Sun, 17 Feb 2008 01:31:56 -0800 (PST), Andrew FPGA
<andrew.newsgroup@gmail.com> wrote:

>Hi there, >I'm trying to simulate the adjacent channel(ACI) performance of a >continuous phase modulation(CPM) scheme so that I can determine if my >modulator/demodulator meets a particular ETSI performance >specification. I'm not sure if my simulation should include AWGN as >well as the ACI? Ultimately I'm trying to chose a combination of CPM >modulation parameters (alphabet size, pulse shape, pulse length, >modulation index) that meet a specific power spectral density channel >mask, SNR performance, and adjacent channel interference performance. > >My setup: >data source -> CPM modulator -> AWGN -> CPM demodulator -> error >counter > >By changing the AWGN eb/no and measuring the BER I can generate BER vs >eb/no curves that match the literature.
That's a good indication that things are working well, and a good reference for the ACI performance.
>For ACI, the ETSI specifcation says "For a carrier to interference >ratio of 0dB, a maximum degradation in the Receive input Signal Level >threshold(RSL) of 1 dB is allowed".
Since RSL is the metric by which ACI performance is judged, that definition is probably the key thing.
>The RSL spec essentially says the receiver must achieve a minimum BER >of 10^-6 at an eb/no of 15dB.
Cool. My interpretation of that would be that you look on your BER vs Eb/No curve and see what Eb/No you're achieving at Pe = 10e-6 (should be 15dB or lower), then add one dB to that value and see what the BER is at that value. That's the BER you're shooting for with 0dB ACI at the same AWGN level that produced Pe = 10e-6. Armed with a BER vs Eb/No curve, you can test with 0dB ACI anywhere along the curve and see if you've degraded 1dB. I don't think you need to prove that to pass, nor would you necessarily be able to in any system. Proving it just at one point (i.e., Eb/No = 15dB, then turn on the ACI) is usually adequate in most standards if that's how it's specified.
>In order to test the ACI performance, the first thing I did was create >ACI of the same power as the orignal signal(0dB) by duplicating the >data source and CPM modulator, and frequency shifting the result so it >lies in the adjacent channel. I then add the ACI to the original >signal after the AWGN in the above diagram.
My only suggestion there would be to make sure that the ACI signal is independently modulated, i.e., different data than the desired signal at a minimum. Making the symbol clock phases indpendent can often be useful (and revealing) as well. The idea is just to make them as uncorrelated as possible.
>How do I check that the RSL(eb/no) degradation is less than 1dB? Do I >reduce the AWGN by 1dB and check that the BER is less than or equal to >the BER without the ACI?
As mentioned, I'd keep the AWGN the same and just check that the BER didn't degrade more than it would be with 1dB more AWGN.
>Should I even have AWGN in my ACI simulation? - I notice that as the >ACI gets stronger, the AWGN has less effect on BER, and at a certain >point the AWGN has no effect on BER - I suppose the ACI is >dominating.
Yes, the ACI is starting to dominate. So I think the test only makes sense at the point stated in the RSL definition. Since that's the definition used to measure ACI performance, I'd think testing at that BER (i.e., Pe = 10e-6) and then checking that the BER is not worse than it would be with 1dB more AWGN would meet the criterion. I suppose another way to interpret it would be to measure the constellation SNR at Pe = 10e-6, and then turn on the ACI and measure it again. If it's not degraded by more than 1dB, it's a pass. If this is something that ultimately needs to pass certification or some sort of compliance, it would be important to find out how that particular test is done to verify cert/compliance. Eric Jacobsen Minister of Algorithms Abineau Communications http://www.ericjacobsen.org
Thanks Eric and John. I think you have made me realise a good thing is
to perform the measurement at the Eb/No that gives 10-6 BER.

The reason I was thinking I wasn't doing it quite right was that at
increased levels of ACI the AWGN had no effect - so basically for that
CPM scheme the test is a complete fail - even with no AWGN a BER of
10-6 cannot ever be achieved with that level of ACI.

The channel spacing cannot be increased of course, so I guess I need
to reduce the amount of ACI present in the channel bandwidth. I could
achieve this by narrowing the transimitted spectrum by selecting a
different set of CPM paramaters (e.g. longer phase pulse). Or I
suppose I could filter the received signal more sharply - in fact I
currently have no filtering in my simulation at all. Ignoring
implementation cost, I suppose the downside of filtering the rx signal
is that the tails of the inband CPM signal are being attenuated as
well as attenuating the ACI.

I feel like I'm missing some info here - surely the IF stage will
provide some ACI rejection also? The rx signal has to be bandlimited
at some point, which presumably provides some ACI rejection also. My
world so far is the digital rx implementation - maybe I need to start
looking and understanding a bit beyond that.....

Cheers
Andrew
On Feb 17, 4:13 pm, andrew.brid...@gmail.com wrote:
> Thanks Eric and John. I think you have made me realise a good thing is > to perform the measurement at the Eb/No that gives 10-6 BER. > > The reason I was thinking I wasn't doing it quite right was that at > increased levels of ACI the AWGN had no effect - so basically for that > CPM scheme the test is a complete fail - even with no AWGN a BER of > 10-6 cannot ever be achieved with that level of ACI. > > The channel spacing cannot be increased of course, so I guess I need > to reduce the amount of ACI present in the channel bandwidth. I could > achieve this by narrowing the transimitted spectrum by selecting a > different set of CPM paramaters (e.g. longer phase pulse). Or I > suppose I could filter the received signal more sharply - in fact I > currently have no filtering in my simulation at all. Ignoring > implementation cost, I suppose the downside of filtering the rx signal > is that the tails of the inband CPM signal are being attenuated as > well as attenuating the ACI. > > I feel like I'm missing some info here - surely the IF stage will > provide some ACI rejection also? The rx signal has to be bandlimited > at some point, which presumably provides some ACI rejection also. My > world so far is the digital rx implementation - maybe I need to start > looking and understanding a bit beyond that..... > > Cheers > Andrew
There is no substitute for a good analog IF filter. I think it should be included in the sim. John
On Sun, 17 Feb 2008 14:13:08 -0800 (PST), John <sampson164@gmail.com>
wrote:

>On Feb 17, 4:13 pm, andrew.brid...@gmail.com wrote: >> Thanks Eric and John. I think you have made me realise a good thing is >> to perform the measurement at the Eb/No that gives 10-6 BER. >> >> The reason I was thinking I wasn't doing it quite right was that at >> increased levels of ACI the AWGN had no effect - so basically for that >> CPM scheme the test is a complete fail - even with no AWGN a BER of >> 10-6 cannot ever be achieved with that level of ACI. >> >> The channel spacing cannot be increased of course, so I guess I need >> to reduce the amount of ACI present in the channel bandwidth. I could >> achieve this by narrowing the transimitted spectrum by selecting a >> different set of CPM paramaters (e.g. longer phase pulse). Or I >> suppose I could filter the received signal more sharply - in fact I >> currently have no filtering in my simulation at all. Ignoring >> implementation cost, I suppose the downside of filtering the rx signal >> is that the tails of the inband CPM signal are being attenuated as >> well as attenuating the ACI. >> >> I feel like I'm missing some info here - surely the IF stage will >> provide some ACI rejection also? The rx signal has to be bandlimited >> at some point, which presumably provides some ACI rejection also. My >> world so far is the digital rx implementation - maybe I need to start >> looking and understanding a bit beyond that..... >> >> Cheers >> Andrew > >There is no substitute for a good analog IF filter. I think it should >be included in the sim. > >John
AI generally agree with John on the use of an IF filter. An exception might be if the tuning is done completely digitally, i.e., post ADC, which would mean that the analog processing would have to include all of the tuning bandwidth. Even in such a case there should be some filtering done to isolate the desired signal and reject some of the ACI. At least, if you need to. It's often tough to meet ACI specs without some sort of IF or channel selection filtering, but if you can meet all specs without it, then there's no need. Eric Jacobsen Minister of Algorithms Abineau Communications http://www.ericjacobsen.org
On Feb 18, 12:33&#2013266080;am, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
> On Sun, 17 Feb 2008 14:13:08 -0800 (PST), John <sampson...@gmail.com> > wrote: > > > > > > >On Feb 17, 4:13 pm, andrew.brid...@gmail.com wrote: > >> Thanks Eric and John. I think you have made me realise a good thing is > >> to perform the measurement at the Eb/No that gives 10-6 BER. > > >> The reason I was thinking I wasn't doing it quite right was that at > >> increased levels of ACI the AWGN had no effect - so basically for that > >> CPM scheme the test is a complete fail - even with no AWGN a BER of > >> 10-6 cannot ever be achieved with that level of ACI. > > >> The channel spacing cannot be increased of course, so I guess I need > >> to reduce the amount of ACI present in the channel bandwidth. I could > >> achieve this by narrowing the transimitted spectrum by selecting a > >> different set of CPM paramaters (e.g. longer phase pulse). Or I > >> suppose I could filter the received signal more sharply - in fact I > >> currently have no filtering in my simulation at all. Ignoring > >> implementation cost, I suppose the downside of filtering the rx signal > >> is that the tails of the inband CPM signal are being attenuated as > >> well as attenuating the ACI. > > >> I feel like I'm missing some info here - surely the IF stage will > >> provide some ACI rejection also? The rx signal has to be bandlimited > >> at some point, which presumably provides some ACI rejection also. My > >> world so far is the digital rx implementation - maybe I need to start > >> looking and understanding a bit beyond that..... > > >> Cheers > >> Andrew > > >There is no substitute for a good analog IF filter. I think it should > >be included in the sim. > > >John > > AI generally agree with John on the use of an IF filter. &#2013266080; An > exception might be if the tuning is done completely digitally, i.e., > post ADC, which would mean that the analog processing would have to > include all of the tuning bandwidth. &#2013266080;Even in such a case there should > be some filtering done to isolate the desired signal and reject some > of the ACI. > > At least, if you need to. &#2013266080; It's often tough to meet ACI specs without > some sort of IF or channel selection filtering, but if you can meet > all specs without it, then there's no need. > > Eric Jacobsen > Minister of Algorithms > Abineau Communicationshttp://www.ericjacobsen.org- Hide quoted text - > > - Show quoted text -
Andrew, When setting the dynamic range of the A/D , don't forget to include the power that may be in one or both of the adjacent channels as well as the desird channel. If the additional power in the unwanted channels causes the A/D to clip, you will have trouble.. Mark
Hi all,
Thanks for all the comments. It turns out the IF stages don't provide
any significant ACI rejection. There is however, a digital ACI reject
filter which I needed to include in the sim.

Cheers
Andrew
Hi All,
Thanks for your comments. It turns out the IF stages don't provide any
significant close in ACI rejection. However, there is a digital ACI
filter I needed to include in my sims.

Cheers
Andrew