Digital Upconversion using the baseband data(IQ) files

On Mon, 25 May 2015 15:02:26 -0500, "kaz" <37480@DSPRelated> wrote:

>>kaz <37480@DSPRelated> wrote:
>>
>>>if your baseband is of width 0~B MHz and you target to move it to
>>+70Mhz
>>>centre then make sure the sampling rate at mixer is > 2*(70 + B) Msps.
>>
>>Actually it can be made to work with the mixer implemented at any
>>sample rate > 2*B Msps.  This is sometimes called subsampling.  But then
>>
>>you would definitely need a post-filter, and *its* sample rate (if it
>>is a digital filter) would have to be > 2*(70+B) Msps.
>>
>>Steve
>
>But the target is 70Mhz absolute
>
>Kaz


Yes, if you're reasonably careful you don't need any additional
filtering.   If you're not careful, and there's an analog IF filter,
you probably still don't need any additional digital filtering.

Since the sample rate at the output of the mixer is generally high,
I've always avoided digital filtering there to keep the complexity
low.   This has been for oversampled or subsampled applications.  So
far I've never had to do it, other than for something like a DAC droop
correction.


Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

>kaz <37480@DSPRelated> wrote:
>
>>>Actually it can be made to work with the mixer implemented at any
>>>sample rate > 2*B Msps.  This is sometimes called subsampling.  But
>then
>
>Oops.  4*B Msps, since the bandwidth of the signal being constructed
>is 2*B MHz, centered at 70 MHz.
>
>>>you would definitely need a post-filter, and *its* sample rate (if it
>>>is a digital filter) would have to be > 2*(70+B) Msps.
>
>>But the target is 70Mhz absolute
>
>So long as the Nyquist criterion is met, the signal will contain
>the desired component centered at 70 MHz, along with other aliases.  
>Higher sample rates place those aliases further away from the component
>of interest, and therefore easier to filter out.
>
>Steve

Steve,

As my mind is limited to FPGA boundary I am a bit surprised at this
strategy. Is it really that common in "soft" dsp practice. You are going
the complicated way just to have lower sampling rate at mixing. 
I assume you will generate say 35Mhz and hence its multiples e.g. 70Mhz
then you will upsample baseband to that required by 35Mhz mixer (instead
of 70) then you create problem of aliasing then go to a higher sampling
rate to filter out the trouble created? or do you expect the analogue
engineer to sort it out. 

Kaz
---------------------------------------
Posted through http://www.DSPRelated.com

kaz <37480@DSPRelated> replies to my post:

>> 4*B Msps, since the bandwidth of the signal being constructed
>a >is 2*B MHz, centered at 70 MHz.

>>>>you would definitely need a post-filter, and *its* sample rate (if it
>>>>is a digital filter) would have to be > 2*(70+B) Msps.

>>>But the target is 70Mhz absolute

>>So long as the Nyquist criterion is met, the signal will contain
>>the desired component centered at 70 MHz, along with other aliases.  
>>Higher sample rates place those aliases further away from the component
>>of interest, and therefore easier to filter out.

>yes you can do it that way 

Okay

>but if I say anything like that in my field (FPGA DSP) I will lose my 
>job instantly. 

(Possibly a good thing, if you're working for people without a grasp
of the basics of sampling and who fire someone on a hair-trigger.)

>We do undersampling at adc level
>of IF signal but in the digits we don't. FPGA target minimum resource and
>minimum modules. Soft DSP can play with equations as they wish. It looks
>like there is a big big difference in the prespectives.

I was not recommending subsampling in implementing this mixer; I was 
instead replying to your statement which I perceived was worded to state 
that it was mathematically necessary to oversample above 140 Ms/sec.

(Perhaps you weren't saying this and it was just my perception.)


Steve

Eric Jacobsen <eric.jacobsen@ieee.org> wrote:

>On Mon, 25 May 2015 15:02:26 -0500, "kaz" <37480@DSPRelated> wrote:

>>>>if your baseband is of width 0~B MHz and you target to move it to
>>>+70Mhz
>>>>centre then make sure the sampling rate at mixer is > 2*(70 + B) Msps.

>Yes, if you're reasonably careful you don't need any additional
>filtering.   If you're not careful, and there's an analog IF filter,
>you probably still don't need any additional digital filtering.

I'm going to say that you always need a filter after this mixer.  It
may be the filter is in the analog chain; and it might not even
be an intentional filter, just a naturally-occuring filtering
from parasitics, but it's stil here.

>Since the sample rate at the output of the mixer is generally high,
>I've always avoided digital filtering there to keep the complexity
>low.   This has been for oversampled or subsampled applications.  So
>far I've never had to do it, other than for something like a DAC droop
>correction.

So let's say you sample at 210 Ms/sec, a convenient number for this
problem.  There will be aliases at 140 MHz and 280 MHz, and so forth.
The only way these don't show up in your system output is they are
being filtered out.  Maybe a sinx/x filter inherenet in your
DAC is enough, maybe your next IF stage doesn't possibly have response at
140 MHz; but it still constitutes a necessary filtering.

Particularly if you are in a regulatory environment, you might need
an explicit filter.

Steve

>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>On Mon, 25 May 2015 15:02:26 -0500, "kaz" <37480@DSPRelated> wrote:
>
>>>>>if your baseband is of width 0~B MHz and you target to move it to
>>>>+70Mhz
>>>>>centre then make sure the sampling rate at mixer is > 2*(70 + B)
>Msps.
>
>>Yes, if you're reasonably careful you don't need any additional
>>filtering.   If you're not careful, and there's an analog IF filter,
>>you probably still don't need any additional digital filtering.
>
>I'm going to say that you always need a filter after this mixer.  It
>may be the filter is in the analog chain; and it might not even
>be an intentional filter, just a naturally-occuring filtering
>from parasitics, but it's stil here.

It looks like we discuss issues at system level rather than in the context
of a beginner's post asking about mixing/upsampling. This is fair enough
but in my work environment we have boards with almost fixed DAC/ADC
circuitry and most of changes are done at fpga level. In other words I as
digital designer will leave LPF of say DAC images to the existing analogue
design and only focus on FPGA domain. Hence I don't need any filter after
mixing. But the system needs it after DAC. As to sinx/x correction etc (I
can add equaliser, DPD...) are realy irrelevant to the post but do add
some practical stuff.

The positive thing I started to have better understanding of various posts
here after so many years.

Kaz



---------------------------------------
Posted through http://www.DSPRelated.com

On Mon, 25 May 2015 21:08:56 +0000 (UTC), spope33@speedymail.org
(Steve Pope) wrote:

>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>On Mon, 25 May 2015 15:02:26 -0500, "kaz" <37480@DSPRelated> wrote:
>
>>>>>if your baseband is of width 0~B MHz and you target to move it to
>>>>+70Mhz
>>>>>centre then make sure the sampling rate at mixer is > 2*(70 + B) Msps.
>
>>Yes, if you're reasonably careful you don't need any additional
>>filtering.   If you're not careful, and there's an analog IF filter,
>>you probably still don't need any additional digital filtering.
>
>I'm going to say that you always need a filter after this mixer.  It
>may be the filter is in the analog chain; and it might not even
>be an intentional filter, just a naturally-occuring filtering
>from parasitics, but it's stil here.
>
>>Since the sample rate at the output of the mixer is generally high,
>>I've always avoided digital filtering there to keep the complexity
>>low.   This has been for oversampled or subsampled applications.  So
>>far I've never had to do it, other than for something like a DAC droop
>>correction.
>
>So let's say you sample at 210 Ms/sec, a convenient number for this
>problem.  There will be aliases at 140 MHz and 280 MHz, and so forth.
>The only way these don't show up in your system output is they are
>being filtered out.  Maybe a sinx/x filter inherenet in your
>DAC is enough, maybe your next IF stage doesn't possibly have response at
>140 MHz; but it still constitutes a necessary filtering.
>
>Particularly if you are in a regulatory environment, you might need
>an explicit filter.
>
>Steve

Yup.   The images in the repeating Nyquist zones need to be considered
depending on the application.   That'll be true regardless of the
system and is usually in the domain of the reconstruction filter.

Usually if somebody is putting an IF at 70MHz there's a lot more
filtering before it gets where it's going.

But, as said before, there is no need to put a digital filter after
the mixer.   Even if you did, you'd still need to deal with the images
in the repeating Nyquist zones, so that's not part of any tradeoff in
the digital domain.



Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

kaz <37480@DSPRelated> replies to my post,

>>So long as the Nyquist criterion is met, the signal will contain
>>the desired component centered at 70 MHz, along with other aliases.  
>>Higher sample rates place those aliases further away from the component
>>of interest, and therefore easier to filter out.

>As my mind is limited to FPGA boundary I am a bit surprised at this
>strategy. Is it really that common in "soft" dsp practice. You are going
>the complicated way just to have lower sampling rate at mixing. 

You said that the OP "needs to" sample at the higher rate in the
mixer; I am simply stating this is not absolutely true.

>I assume you will generate say 35Mhz and hence its multiples e.g. 70Mhz
>then you will upsample baseband to that required by 35Mhz mixer (instead
>of 70) then you create problem of aliasing then go to a higher sampling
>rate to filter out the trouble created? or do you expect the analogue
>engineer to sort it out. 

I can think of reasons you'd want to subsample.  Say the signal bandwidth
is low, say 50 KHz.  After the subsampled digital mixer, you could 
implment a sigma-delta modulator, at a clock rate in the 2 to 4 MHz range, 
that would emit a pulse-like signal that feeds into the analog domain, 
and after that have a narrow, high-Q analog filter with a 70 MHz center 
frequency from which the signal of interest could be recovered and 
amplified.

That may not sound very classy, but I can tell you it's extremely
inexpensive -- all of your digital processing is at a low sample
rate, and the analog circuits can be made to be cheap.

You'd consider something like that for a high-volume, low-cost
part (admittedly outside of the range of the OP's description of their
problem).

Steve

On Monday, May 25, 2015 at 3:54:15 AM UTC-4, ikram...@gmail.com wrote:
> Hi all,
>      i want to upconvert the baseband data (complex data). i have code for upconverting but struggling to implement interpolater (resampler) before upconverter and i need to implement a filter after the upconverter. so can any one help me to implement both resampler and filter . Iam trying to upconvert a baseband to 70Mhz IF .
> 
> 
> 
> Thanks in advance.

What do you do for a living?

Dirk Bell

On Tue, 26 May 2015 02:05:12 +0000 (UTC), spope33@speedymail.org
(Steve Pope) wrote:

>kaz <37480@DSPRelated> replies to my post,
>
>>>So long as the Nyquist criterion is met, the signal will contain
>>>the desired component centered at 70 MHz, along with other aliases.  
>>>Higher sample rates place those aliases further away from the component
>>>of interest, and therefore easier to filter out.
>
>>As my mind is limited to FPGA boundary I am a bit surprised at this
>>strategy. Is it really that common in "soft" dsp practice. You are going
>>the complicated way just to have lower sampling rate at mixing. 
>
>You said that the OP "needs to" sample at the higher rate in the
>mixer; I am simply stating this is not absolutely true.
>
>>I assume you will generate say 35Mhz and hence its multiples e.g. 70Mhz
>>then you will upsample baseband to that required by 35Mhz mixer (instead
>>of 70) then you create problem of aliasing then go to a higher sampling
>>rate to filter out the trouble created? or do you expect the analogue
>>engineer to sort it out. 
>
>I can think of reasons you'd want to subsample.  Say the signal bandwidth
>is low, say 50 KHz.  After the subsampled digital mixer, you could 
>implment a sigma-delta modulator, at a clock rate in the 2 to 4 MHz range, 
>that would emit a pulse-like signal that feeds into the analog domain, 
>and after that have a narrow, high-Q analog filter with a 70 MHz center 
>frequency from which the signal of interest could be recovered and 
>amplified.
>
>That may not sound very classy, but I can tell you it's extremely
>inexpensive -- all of your digital processing is at a low sample
>rate, and the analog circuits can be made to be cheap.
>
>You'd consider something like that for a high-volume, low-cost
>part (admittedly outside of the range of the OP's description of their
>problem).
>
>Steve

Those are all very good reasons to do that, and we've designed
products that way in the past.   If you can find an economical DAC
that has good enough output bandwidth to support subsampling, you can
potentially do away with a more complex (and expensive) AFE and just
digitize directly to IF.    Years ago we used that to get a 70MHz IF
with a 100MHz DAC and have also done 140 MHz IF the same way.   It
worked great with very low MER, but required carefully designed DAC
correction filters.   

These days DACs with higher sampling rates are not hard to come by,
but as you point out that means that the digital stuff has to run much
faster, which can also add significant cost depending on the
implementation.

Good tricks to know, in any case.


Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

On Mon, 25 May 2015 00:54:11 -0700 (PDT), ikram998563@gmail.com wrote:

>Hi all,
>     i want to upconvert the baseband data (complex data). i have code for upconverting but struggling to implement interpolater (resampler) before upconverter and i need to implement a filter after the upconverter. so can any one help me to implement both resampler and filter . Iam trying to upconvert a baseband to 70Mhz IF .
>
>Thanks in advance.

Hello ikram998563,
   After reading all the posts in this thread, 
do you now have the answer to your question?

[-Rick-]