Complex Baseband

Howdy,
For FM/AM demod, you require a complex baseband. I am downconverting
the IF from 1MHz to baseband by multiplying the input stream (12 bits)
from the ADC with Sin and Cosine outputs of a DDS and therefore,
generating I, Q
I am using standard 18x18 signed multipliers in Xilinx to do the
mixing. Is this right or should I use a complex multiplier?
The reason why I am asking this question is because, I had implemented
the whole datapath (Downconversion, Rate decimation, Compensating FIR)
and verified it on the bench and then implemented CORDIC for getting
FM, which never worked.
So instead I tried to do AM demod by not using CORDIC but (Sq(I) +
Sq(Q)) using standard 18x18 signed multipliers and still no AM demod
which is supposed to be simple.
Then I took the CIC and CFIR filters out and tried to do demod
straight after the mixing, still nothing.
So I'm kinda stuck here, any help will be appreciated
Thanks
-M
p.s. I don't have access to Matlab and System generator, but I'm
buying them next week. I figured, doing digital radio design without
modeling tools is like digging your own grave!!!

On Mar 29, 7:09 pm, "morpheus" <saurs...@gmail.com> wrote:
> Howdy,
> For FM/AM demod, you require a complex baseband. I am downconverting
> the IF from 1MHz to baseband by multiplying the input stream (12 bits)
> from the ADC with Sin and Cosine outputs of a DDS and therefore,
> generating I, Q
> I am using standard 18x18 signed multipliers in Xilinx to do the
> mixing. Is this right or should I use a complex multiplier?
> The reason why I am asking this question is because, I had implemented
> the whole datapath (Downconversion, Rate decimation, Compensating FIR)
> and verified it on the bench and then implemented CORDIC for getting
> FM, which never worked.
> So instead I tried to do AM demod by not using CORDIC but (Sq(I) +
> Sq(Q)) using standard 18x18 signed multipliers and still no AM demod
> which is supposed to be simple.
> Then I took the CIC and CFIR filters out and tried to do demod
> straight after the mixing, still nothing.
> So I'm kinda stuck here, any help will be appreciated
> Thanks
> -M
> p.s. I don't have access to Matlab and System generator, but I'm
> buying them next week. I figured, doing digital radio design without
> modeling tools is like digging your own grave!!!



Multipling a real times a complex only needs two real multiplies, so
you do not need a complex multiplier. What you do need to keep an eye
on is where you centered your complex signal, and does this match what
the downstream processing is expecting.  Did you shift your signal to
be at DC to fs/2, or from -fs/4 to fs/f? Did you filter and decimate
after mixing, or leave it oversampled?  There are multiple ways to do
this, and if the different parts do not agree, you get garbage out.

Here is a starting point for some further explanation:

http://en.wikipedia.org/wiki/Analytic_signal

Regards,

John McCaskill
www.fastertechnology.com

Hey John,
I have centered the signal between DC and fs/2. My problem is that
even though the I and Q channels are getting mixed correctly (verified
by offsetting the carrier frequency and watching the modulation move
by the offset value around DC on a spectrum analyzer), the demod is
not working, sq(I) + sq(Q) seems like a dead output....i checked the
synthesis report and post P&R simulation and it seems to be working in
simulation.
Bizarro!!!
-M

On Mar 29, 6:08 pm, "morpheus" <saurs...@gmail.com> wrote:
> Hey John,
> I have centered the signal between DC and fs/2. My problem is that
> even though the I and Q channels are getting mixed correctly (verified
> by offsetting the carrier frequency and watching the modulation move
> by the offset value around DC on a spectrum analyzer), the demod is
> not working, sq(I) + sq(Q) seems like a dead output....i checked the
> synthesis report and post P&R simulation and it seems to be working in
> simulation.
> Bizarro!!!
> -M

When I say its "working" in simulation, I mean that through LabView I
generated a FM modulated signal sampled at 50MSPS (same as my system)
and used that as the input test vector to my design.
I am doing functional verification using ModelSim and I can see that
I*I + Q*Q outputs being computed, but they don't seem to be present in
hardware.
I am thinking there is something weird going on system wise that is
leading to this anomaly, maybo some harmonics that are leading to
cancellations or something, because I just cannot understand why the
I*I multipliers are not working

"morpheus" <saurster@gmail.com> wrote in message 
news:1175226348.769913.256740@r56g2000hsd.googlegroups.com...
>
> When I say its "working" in simulation, I mean that through LabView I
> generated a FM modulated signal sampled at 50MSPS (same as my system)
> and used that as the input test vector to my design.
> I am doing functional verification using ModelSim and I can see that
> I*I + Q*Q outputs being computed, but they don't seem to be present in
> hardware.
> I am thinking there is something weird going on system wise that is
> leading to this anomaly, maybo some harmonics that are leading to
> cancellations or something, because I just cannot understand why the
> I*I multipliers are not working
>
Chipscope? 

On 30 Mrz., 02:09, "morpheus" <saurs...@gmail.com> wrote:

> p.s. I don't have access to Matlab and System generator, but I'm
> buying them next week. I figured, doing digital radio design without
> modeling tools is like digging your own grave!!!

But you have access to Octave and Scilab.
Both free and powerful.

Kojla Sulimma

On Mar 30, 4:47 am, "comp.arch.fpga" <ksuli...@googlemail.com> wrote:
> On 30 Mrz., 02:09, "morpheus" <saurs...@gmail.com> wrote:
>
> > p.s. I don't have access to Matlab and System generator, but I'm
> > buying them next week. I figured, doing digital radio design without
> > modeling tools is like digging your own grave!!!
>
> But you have access to Octave and Scilab.
> Both free and powerful.
>
> Kojla Sulimma

Well, honestly I didnt know about these tools, I don't have chipscope
either. I did get an A/D board from LT which has a daughter board for
data collection. This is a cool boar, it attaches to the A/D board and
has a FPGA on it (with a big-ass heat sink on the FPGA). Anyways, they
have software to log the A/D data and compute other parameters like
SINAD/ SNR, etc.
I am planning to use this board to collect data and use that as input
to my simulation
cheers

"morpheus" <saurster@gmail.com> wrote in message 
news:1175267526.611102.281320@p15g2000hsd.googlegroups.com...
>
> Well, honestly I didnt know about these tools, I don't have chipscope
> either. I did get an A/D board from LT which has a daughter board for
> data collection. This is a cool boar, it attaches to the A/D board and
> has a FPGA on it (with a big-ass heat sink on the FPGA). Anyways, they
> have software to log the A/D data and compute other parameters like
> SINAD/ SNR, etc.
> I am planning to use this board to collect data and use that as input
> to my simulation
> cheers
>
Dear Morpheus,
http://www.xilinx.com/ise/optional_prod/cspro.htm
Click on 'evaluate'.
HTH, Syms.
p.s. I've got one of those LT boards too. V.cool. 

morpheus wrote:
> Howdy,
> For FM/AM demod, you require a complex baseband. I am downconverting
> the IF from 1MHz to baseband by multiplying the input stream (12 bits)
> from the ADC with Sin and Cosine outputs of a DDS and therefore,
> generating I, Q
> I am using standard 18x18 signed multipliers in Xilinx to do the
> mixing. Is this right or should I use a complex multiplier?
> The reason why I am asking this question is because, I had implemented
> the whole datapath (Downconversion, Rate decimation, Compensating FIR)
> and verified it on the bench and then implemented CORDIC for getting
> FM, which never worked.
> So instead I tried to do AM demod by not using CORDIC but (Sq(I) +
> Sq(Q)) using standard 18x18 signed multipliers and still no AM demod
> which is supposed to be simple.
> Then I took the CIC and CFIR filters out and tried to do demod
> straight after the mixing, still nothing.
> So I'm kinda stuck here, any help will be appreciated
> Thanks
> -M
> p.s. I don't have access to Matlab and System generator, but I'm
> buying them next week. I figured, doing digital radio design without
> modeling tools is like digging your own grave!!!
> 


If your IF is complex, then you'll need a complex multiplier to multiply 
your complex IF by the complex LO (cos + j*sin).  If, on the other hand, 
your IF is real-only (which I suspect it is), then you multiply the real 
IF by the complex LO.  That uses a pair of real-only multipliers (note, 
this is the same as using a complex multiplier and tying the Q part of 
the IF input to 0).

Is your LO tuned properly to get a signal to baseband?

And yes, you should have Matlab or another similar tool to let you work 
on the algorithmic implementation before you start the actual design, as 
well as to model the design for verification purposes.  You don't need 
System Generator to turn out DSP designs for FPGAs, in fact I find it 
gets in the way more often than not for the high speed stuff I am 
usually involved with.

Ray,
Your suggestion makes sense. I had implemented real only multiplier as
my IF is real.
Regarding the output of the envelope detector, I think, due to
truncation at various stages, I have had to drive the ADC, DAC really
high to get some sort of an output. I need to rethink my truncation
strategy at each stage as I am losing tons of dynamic range and this
would be helped if I have a system modeling tool.
Bless the good people at MathWorks.
Thanks for all the comments
-M