Question on GSM multipath channel modelling and performance measurement

Hi folks,

I just implemented a MLSE software module for GSM signal demodulation and
now want to test its performance by following 3GPP SPEC (TS45.005 section
6).

The problem I am facing now is how to apply the multipath channel
specified in the above spec. The Annex C of TS45.005 gives many tables in
terms of Relative delay time and average relative power. My question is how
to make use of them?

Take the Rural Area (RAx) table in Annex C for example. Can I simply put
the relative power values as FIR tap coefficients and then pass my signal
through this FIR filter to get a multipath-distorted signal? In this case,
I must also sample my input signal with the same time interval as the
Relative Time. Is my thinking correct?

I also saw in some articles that receiver speed is also used in
calculating multipath channel, e.g., EQ50, RA50 etc. Can anyone give some
hints on how to do this?

in brief, my question is how to model a correct multipath channel based on
the information provided in TS45.005 Annex C. Please help.

thank you


Richard


 

richard_zhang <hardheartcn@yahoo.com.cn> wrote:

> I just implemented a MLSE software module for GSM signal
> demodulation and now want to test its performance by following
> 3GPP SPEC (TS45.005 section 6).

> The problem I am facing now is how to apply the multipath
> channel specified in the above spec. The Annex C of TS45.005
> gives many tables in terms of Relative delay time and average
> relative power. My question is how to make use of them?

> Take the Rural Area (RAx) table in Annex C for example. Can I
> simply put the relative power values as FIR tap coefficients
> and then pass my signal through this FIR filter to get a
> multipath-distorted signal? In this case, I must also sample
> my input signal with the same time interval as the Relative
> Time. Is my thinking correct?

> I also saw in some articles that receiver speed is also used
> in calculating multipath channel, e.g., EQ50, RA50 etc. Can
> anyone give some hints on how to do this?

> in brief, my question is how to model a correct multipath
> channel based on the information provided in TS45.005 Annex
> C. Please help.  thank you

It somewhat depends upon your purpose in doing these
simulation tests.  Assuming you are doing something along
the lines of debugging your modem implementation, and obtaining a 
believable performance curve that is generally correct, you can 
do what you describe -- perform any sample-rate conversion,
apply an FIR filter to the I/Q signal, the magnitude of the taps 
corresponds to the power-delay values (i.e. magnitude equals
square root of power), and the phase of these taps being
uniform random in the interval 0 to 2 * pi.

If you are doing actual conformance testing, then this is
probably not sufficient.  Almost all GSM development projects
would rely on a commercially available conformance test suite 
since you could not otherwise sell the design into a handset.

Steve

>richard_zhang <hardheartcn@yahoo.com.cn> wrote:
>
>> I just implemented a MLSE software module for GSM signal
>> demodulation and now want to test its performance by following
>> 3GPP SPEC (TS45.005 section 6).
>
>> The problem I am facing now is how to apply the multipath
>> channel specified in the above spec. The Annex C of TS45.005
>> gives many tables in terms of Relative delay time and average
>> relative power. My question is how to make use of them?
>
>> Take the Rural Area (RAx) table in Annex C for example. Can I
>> simply put the relative power values as FIR tap coefficients
>> and then pass my signal through this FIR filter to get a
>> multipath-distorted signal? In this case, I must also sample
>> my input signal with the same time interval as the Relative
>> Time. Is my thinking correct?
>
>> I also saw in some articles that receiver speed is also used
>> in calculating multipath channel, e.g., EQ50, RA50 etc. Can
>> anyone give some hints on how to do this?
>
>> in brief, my question is how to model a correct multipath
>> channel based on the information provided in TS45.005 Annex
>> C. Please help.  thank you
>
>It somewhat depends upon your purpose in doing these
>simulation tests.  Assuming you are doing something along
>the lines of debugging your modem implementation, and obtaining a 
>believable performance curve that is generally correct, you can 
>do what you describe -- perform any sample-rate conversion,
>apply an FIR filter to the I/Q signal, the magnitude of the taps 
>corresponds to the power-delay values (i.e. magnitude equals
>square root of power), and the phase of these taps being
>uniform random in the interval 0 to 2 * pi.
>
>If you are doing actual conformance testing, then this is
>probably not sufficient.  Almost all GSM development projects
>would rely on a commercially available conformance test suite 
>since you could not otherwise sell the design into a handset.
>
>Steve
>

Hello Steve,

Thank you for reply.

What I am going to do is to check the BER performance of my MLSE
equalizer/detector, based on matlab simulation only.

In your suggested method, I can use the delay power profile from 3GPP spec
and then apply uniformly distributed phase to each tap, which makes sense
to me. But one question is still confusing me, how to generate a multipath
channel including velocity information? Some articles mention they use a
channel model like EQ50, HT100 and so on, where 50 and 100 represents
receiver velocity in Km/hour. Can you illustrate more?

thank you.


Richard 

richard_zhang <hardheartcn@yahoo.com.cn> wrote:

[snip]

> But one question is still confusing me, how to generate a
> multipath channel including velocity information? Some articles
> mention they use a channel model like EQ50, HT100 and so on,
> where 50 and 100 represents receiver velocity in Km/hour. Can
> you illustrate more?

I haven't myself written a Doppler channel model with a velocity
parameter, so I'm not the best person to help you out here.
Perhaps someone else can comment.

I'm not aware of any publically-available such models for GSM,
but you could probably find something for, say, 802.16m with
a little websearching.

Steve

>richard_zhang <hardheartcn@yahoo.com.cn> wrote:
>
>[snip]
>
>> But one question is still confusing me, how to generate a
>> multipath channel including velocity information? Some articles
>> mention they use a channel model like EQ50, HT100 and so on,
>> where 50 and 100 represents receiver velocity in Km/hour. Can
>> you illustrate more?
>
>I haven't myself written a Doppler channel model with a velocity
>parameter, so I'm not the best person to help you out here.
>Perhaps someone else can comment.
>
>I'm not aware of any publically-available such models for GSM,
>but you could probably find something for, say, 802.16m with
>a little websearching.
>
>Steve
>

OK. Thank you very much and hope somebody else can post some help here.

On Sep 3, 3:35 am, "richard_zhang" <hardhear...@yahoo.com.cn> wrote:
> Hi folks,
>
> I just implemented a MLSE software module for GSM signal demodulation and
> now want to test its performance by following 3GPP SPEC (TS45.005 section
> 6).
>
> The problem I am facing now is how to apply the multipath channel
> specified in the above spec. The Annex C of TS45.005 gives many tables in
> terms of Relative delay time and average relative power. My question is how
> to make use of them?
>
> Take the Rural Area (RAx) table in Annex C for example. Can I simply put
> the relative power values as FIR tap coefficients and then pass my signal
> through this FIR filter to get a multipath-distorted signal? In this case,
> I must also sample my input signal with the same time interval as the
> Relative Time. Is my thinking correct?
>
> I also saw in some articles that receiver speed is also used in
> calculating multipath channel, e.g., EQ50, RA50 etc. Can anyone give some
> hints on how to do this?
>
> in brief, my question is how to model a correct multipath channel based on
> the information provided in TS45.005 Annex C. Please help.
>
> thank you
>
> Richard


Richard,
Most of your questions are answered in Sections C.1 and C.2 of the
spec. Essentially, you convert the speed of the vehicle to the Doppler
frequency
and then put the Doppler frequency in the equation in Section C.2 to
determine the power spectral density of the channel tap. The value in
the table
gives the *relative average power* of each tap. So, for instance, on
the average, tap 2 will be 4 dB lower than tap 1 for RAx. However,
that does not mean
that tap 2 will always be 4 dB lower (which is what will be achieved
by your FIR filter approach). Sometimes, tap 2 will be lower than tap
1 and sometimes higher,
but on the average, it will work out to 4 dB lower. The instantaneous
tap coefficients have to be generated based on a random realization of
a channel that has the
PSD properties as specified in the equation.

Hope that helps.

Thanks,
Dilip.