Problem of Hilbert Transfrom

I am using the PADK board based on C6727 dsp chip and want to implement SSB modulation from a monotone input signal. There's a way (from IEEE magazine)to generate the analytic signal composed of the in-phase signal and the quadrature signal using a pair of FIR filter coeffecients A and B (B is the reverse version of A). In the matlab simulation, when the input singal is of mono frequency, the output in-phase and quadrature signals have a uniform amplitude and a pahse difference of 90 degree which can meet the demands for SSB modulation, but in real implementation using the PADK board, the amplitude of in-phase and quadrature output are not the same, I have tried different FIR filter lengths from 50 to 100, the results are also bad.

So anyone can give me some suggestions or some other better ways to generate the Hilbert transform of the input signal?

Thanks!

_____________________________________ Start a New Thread

Franc-

> I am using the PADK board based on C6727 dsp chip and want to implement SSB
> modulation from a monotone input signal. There's a way (from IEEE magazine)
> to generate the analytic signal composed of the in-phase signal and the
> quadrature signal using a pair of FIR filter coeffecients A and B (B is the
> reverse version of A). In the matlab simulation, when the input singal is
> of mono frequency, the output in-phase and quadrature signals have a
> uniform amplitude and a pahse difference of 90 degree which can meet the
> demands for SSB modulation, but in real implementation using the PADK
> board, the amplitude of in-phase and quadrature output are not the same,
> I have tried different FIR filter lengths from 50 to 100, the results are
> also bad.

Can you clarify "bad"? Can you post some graphs or plots?

Is it your C code running on the C6727 or did you use Simulink to generate the C
code? What is your sampling rate?

The C6727 is 32-bit floating-point so any differences from the MATLAB simulation
should be insignificant. My guess is the code running on the DSP is buggy, or
possibly is too slow and falls out of real-time.

-Jeff

_____________________________________ Start a New Thread

The sampling rate is 192kHz, the C codes is written by myself and is running in CCS3.1, I use the function in the dsp library: DSPF_sp_fir_gen()
for FIR filtering. The coefficients A(for inphase output) and B(for quadrature output, reverse order of A) are generated by MATLAB.

The output result of PADK board viewed from oscilloscope is that the amplitudes of inphase and quadrature are not uniform and changes when the input frequecny varies, only are same for several frequencies. So I am confused about this.

Thanks

I am using the PADK board based on C6727 dsp chip and want to implement SSB modulation from a monotone input signal. There's a way (from IEEE magazine)to generate the analytic signal composed of the in-phase signal and the quadrature signal using a pair of FIR filter coeffecients A and B (B is the reverse version of A). In the matlab simulation, when the input singal is of mono frequency, the output in-phase and quadrature signals have a uniform amplitude and a pahse difference of 90 degree which can meet the demands for SSB modulation, but in real implementation using the PADK board, the amplitude of in-phase and quadrature output are not the same, I have tried different FIR filter lengths from 50 to 100, the results are also bad.
>
>So anyone can give me some suggestions or some other better ways to generate the Hilbert transform of the input signal?
>
>Thanks!
>
>_____________________________________

_____________________________________ Start a New Thread

Franc-

> The sampling rate is 192kHz, the C codes is written by myself and
> is running in CCS3.1, I use the function in the dsp
> library: DSPF_sp_fir_gen()
> for FIR filtering. The coefficients A(for inphase output) and B
> (for quadrature output, reverse order of A) are
> generated by MATLAB.
>
> The output result of PADK board viewed from oscilloscope is that
> the amplitudes of inphase and quadrature are not
> uniform and changes when the input frequecny varies, only are same
> for several frequencies. So I am confused about
> this.

192 kHz is fairly fast, depending on what your DSP code is doing. What is your FIR filter length?

My suggestion is to try a very low sampling rate, for example 8 kHz, and see if the dig scope waveforms look Ok. If
not, then it's probably a DSP code issue. If yes, then start increasing sampling rate until you find the "breaking
point", then identify which DSP code is too slow.

-Jeff

> I am using the PADK board based on C6727 dsp chip and want to implement SSB modulation from a monotone input signal.
> There's a way (from IEEE magazine)to generate the analytic signal composed of the in-phase signal and the quadrature
> signal using a pair of FIR filter coeffecients A and B (B is the reverse version of A). In the matlab simulation, when
> the input singal is of mono frequency, the output in-phase and quadrature signals have a uniform amplitude and a pahse
> difference of 90 degree which can meet the demands for SSB modulation, but in real implementation using the PADK
> board, the amplitude of in-phase and quadrature output are not the same, I have tried different FIR filter lengths
> from 50 to 100, the results are also bad.
>>
>>So anyone can give me some suggestions or some other better ways to generate the Hilbert transform of the input
>> signal?
>>
>>Thanks!

_____________________________________ Start a New Thread

Thank you very much!

the length of the FIR filter is 50. And the sampling rate of the PADK board can be set to 48 kHz, 96 kHz and 192 kHz. I am a novice studying DSP programming for only few months. So how can I lower the sampling rate to 8 kHz, by software program using decimation?

Thanks!

I am using the PADK board based on C6727 dsp chip and want to implement SSB modulation from a monotone input signal. There's a way (from IEEE magazine)to generate the analytic signal composed of the in-phase signal and the quadrature signal using a pair of FIR filter coeffecients A and B (B is the reverse version of A). In the matlab simulation, when the input singal is of mono frequency, the output in-phase and quadrature signals have a uniform amplitude and a pahse difference of 90 degree which can meet the demands for SSB modulation, but in real implementation using the PADK board, the amplitude of in-phase and quadrature output are not the same, I have tried different FIR filter lengths from 50 to 100, the results are also bad.
>
>So anyone can give me some suggestions or some other better ways to generate the Hilbert transform of the input signal?
>
>Thanks!
>
>_____________________________________

_____________________________________ Start a New Thread

Franc-

> the length of the FIR filter is 50. And the sampling rate of the
> PADK board can be set to 48 kHz, 96 kHz and 192 kHz. I am a novice
> studying DSP programming for only few months. So how can I lower
> the sampling rate to 8 kHz, by software program using decimation?

I recall that the minimum sampling rate for the PADK board is 32 kHz. That would be
low enough for the "real-time baseline" tests we are discussing, also you can try
much shorter filters and/or bypass the filters to establish the baseline. If 32 kHz
is accurate, I would say it's surprising that Lyrtech didn't allow sampling rate
selection as low as 8 kHz for a professional audio board.

Do you have a link to the PADK reference guide? I couldn't find one on Lyrtech's
website. If I can see the documentation I might be able to give better advice.

-Jeff

PS. Please don't cut out the text or questions from previous replies. It's supposed
to be fun to answer questions and help people out, but if I have to go back and look
up previous posts it wastes time and makes it less fun. When you do that, you make
it less likely you will get a reply.

> Thanks!
>
> I am using the PADK board based on C6727 dsp chip and want to
> implement SSB modulation from a monotone input signal. There's a
> way (from IEEE magazine)to generate the analytic signal composed
> of the in-phase signal and the quadrature signal using a pair of
> FIR filter coeffecients A and B (B is the reverse version of A).
> In the matlab simulation, when the input singal is of mono
> frequency, the output in-phase and quadrature signals have a
> uniform amplitude and a pahse difference of 90 degree which can
> meet the demands for SSB modulation, but in real implementation
> using the PADK board, the amplitude of in-phase and quadrature
> output are not the same, I have tried different FIR filter
> lengths from 50 to 100, the results are also bad.
> >
> >So anyone can give me some suggestions or some other better ways
> > to generate the Hilbert transform of the input signal?

_____________________________________ Start a New Thread

Wei JI

> Thanks! I've attached the reference guide of PADK board.
>
> The sample rate can be set to 48k,96k and 192kHz, and there
> is a sample rate converter on the board, can I use it to
> change the sample rate?

Are you using the digital audio input or analog? The sample rate converter (TI SRC4192) only applies to the digital
audio input.

The PCM4204 data sheet appears to show that lower sampling rates are supported. Suggest to ask Lyrtech about this.

-Jeff

> (Elfranc)
>
> Laboratory of Communication Acoustics
> Institute of Acoustics, Chinese Academy of Sciences
> Beijing, China
>  
>
> --- 09年9月15日，周二, Jeff Brower 写道：
> 发件人: Jeff Brower
> 主题: Re: [c6x] Re: Problem of Hilbert Transfrom
> 收件人: e...@yahoo.com.cn
> 抄送: c...
> 日期: 2009年9月15日,周二,上午9:51
> Franc-
>
>> the length of the FIR filter is 50. And the sampling rate of the
>> PADK board can be set to 48 kHz, 96 kHz and 192 kHz. I am a novice
>> studying DSP programming for only few months. So how can I lower
>> the sampling rate to 8 kHz, by software program using decimation?
>
> I recall that the minimum sampling rate for the PADK board is 32 kHz.  That would be
> low enough for the "real-time baseline" tests we are discussing, also you can try
> much shorter filters and/or bypass the filters to establish the baseline.  If 32 kHz
> is accurate, I would say it's surprising that Lyrtech didn't allow sampling rate
> selection as low as 8 kHz for a professional audio board.
>
> Do you have a link to the PADK reference guide?  I couldn't find one on Lyrtech's
> website.  If I can see the documentation I might be able to give better advice.
>
> -Jeff
>
> PS. Please don't cut out the text or questions from previous replies.  It's supposed
> to be fun to answer questions and help people out, but if I have to go back and look
> up previous posts it wastes time and makes it less fun.  When you do that, you make
> it less likely you will get a reply.
>
>> Thanks!
>>
>> I am using the PADK board based on C6727 dsp chip and want to
>> implement SSB modulation from a monotone input signal. There's a
>> way (from IEEE magazine)to generate the analytic signal composed
>> of the in-phase signal and the quadrature signal using a pair of
>> FIR filter coeffecients A and B (B is the reverse version of A).
>> In the matlab simulation, when the input singal is of mono
>> frequency, the output in-phase and quadrature signals have a
>> uniform amplitude and a pahse difference of 90 degree which can
>> meet the demands for SSB modulation, but in real implementation
>> using the PADK board, the amplitude of in-phase and quadrature
>> output are not the same, I have tried different FIR filter
>> lengths from 50 to 100, the results are also bad.
>> >
>> >So anyone can give me some suggestions or some other better ways
>> > to generate the Hilbert transform of the input signal?

_____________________________________ Start a New Thread