On Friday, May 12, 2017 at 8:22:46 AM UTC-3, Tauno Voipio wrote:
> On 12.5.17 12:55, Tauno Voipio wrote:
> > On 12.5.17 00:44, danielot@gmail.com wrote
> 
> > A question: What kind of filtering do you have in the RF front end
> > before sampling?
> 
> I made a quick image response calculation. You have image
> responses in addition to the desired 500 MHz:
> 
>   164.9 MHz, 278.4 MHz, 386.6 MHz, 608.2 MHz,
>   721.6 MHz, 886.6 MHz, 829.7 MHz, 943.3 MHz,
> 
> and many others, including direct pass-through of 56.7 MHz.
> 
> In digital signal processing, the images can also be called
> aliases.
> 
> If there is any energy in any, some or all of the images,
> it will come to the ADC output indistinguishable from the
> desired 500 MHz.


Thanks.

The RF front-end is basically the following sequence of components:

LPF: Minicircuits LFCN-530+ 
2x2 switch: RFMD RFSW6223
BPF: TAI-SAW TA1114
Amp: Minicircuits TAMP-72LN 
Attenuator: Minicircuits DAT-31R5-SP+ 
LPF: Minicircuits LFCN-530+ 
Amp: Minicircuits TAMP-72LN 
LPF: Minicircuits LFCN-530+

The signal coming in is composed of very short pulses (aprox. 10 ps Gaussian profile) with considerable energy going up to a few tens GHz. And also we have all harmonics starting from 500 MHz / 864 = ~580 kHz. This is called the revolution frequency, which is the repetition rate of the electron bunches train circuliating inside the vacuum tube, whilst the 500 MHz comes from the 2 ns spacing between the bunches (or pulses, from the signal's point of view).

We've made similar analysis to yours to cope with aliased frequencies. We're happy with the lowpass performance but just recently we discovered the attenuation provided below 400 MHz by the BPF (55 dB) could be slightly better and make a positive impact in noise performance. We may want to add a coaxial HPF module between the RFFE and the ADC when we'll be deploying, maybe Minicircuits SHP-500+.

An unbalance of 0.001 dB between channels means ~1.5 um beam position measurement. Our spec is < 80 nm RMS in 0.1 Hz - 1 kHz bandwidth.

On 12.5.17 12:55, Tauno Voipio wrote:
> On 12.5.17 00:44, danielot@gmail.com wrote

> A question: What kind of filtering do you have in the RF front end
> before sampling?

I made a quick image response calculation. You have image
responses in addition to the desired 500 MHz:

  164.9 MHz, 278.4 MHz, 386.6 MHz, 608.2 MHz,
  721.6 MHz, 886.6 MHz, 829.7 MHz, 943.3 MHz,

and many others, including direct pass-through of 56.7 MHz.

In digital signal processing, the images can also be called
aliases.

If there is any energy in any, some or all of the images,
it will come to the ADC output indistinguishable from the
desired 500 MHz.

-- 

-TV

On 12.5.17 00:44, danielot@gmail.com wrote:
> On Thursday, May 11, 2017 at 3:19:51 PM UTC-3, Steve Pope wrote:
>>
>> Yes, if you have the Hilbert transformer at baseband, then you have
>> additional difficulties in recovering components near DC, relative
>> to having the Hilbert transformer at IF (as described in my last pair
>> of posts).
> 
> I finally made my code working! But then I realized it only works for baseband. I'm just wondering now how would it be possible to do the Hilbert transformer at IF as you said.
> 
> I have the impression the Fig. 5 of the AARL paper you shared is still describing a baseband Hilbert transformer, which had to be implemented as a BPF due to physical limitations. I think it would be impossible to design a LPF with +90 deg at DC+ frequencies and -90 deg at DC- frequencies.
> 
> Does it look right?
> 
>>
>> Generally -- if you're actually building a receiver -- you need
>> an architecture and a frequency plan, and therefore face a decision
>> between direct conversion and low-IF, ahead of your ADC.  For the
>> SSB case these two choices correspond to the two possible placements
>> for the Hilbert transformer described above.
>>
>> If you're just doing a simulation for a proof of concept, then it may
>> not matter yet, and you can simulate it either way.  But if you're
>> pretty sure which way the final design will go, then save yourself
>> some work and go in that direction now.
> 
> Thanks for the advice.
> 
> We have designed a receiver. By undersampling our 500 MHz carrier at ~221 MHz sampling rate (RF/864*383) we get the IF at 56.7 MHz (RF/864*98)). Then we downconvert to baseband digitally in a FPGA using standard AM demodulation using complex NCO. That's the point where I wanted to improve and use SSB demod.
> 
> This is the project: http://www.ohwr.org/projects/bpm/wiki
> 
> Daniel
> 

A question: What kind of filtering do you have in the RF front end
before sampling?

-- 

-TV

On 12.5.17 05:01, Steve Pope wrote:
> Les Cargill  <lcargill99@comcast.com> wrote:
> 
>> That's my understanding - SSB was vaunted back in the CB radio days
>> to have better range, which I assumed was due to better noise immunity.

SSB is much older. It was used already in the 1930's
in carrier telephony, to squeeze more channels in the
same cable.

> It has better range than AM, other things being equal, because you
> are not wasting power transmitting a carrier with no information.
> 
> Double sideband suppessed carrier should be as good as SSB --
> twice the noise bandwidth but a 2x redundant signal -- but you
> get half as many users in a band.

The problem in receiving suppressed-carrier DSB is that you need
to have the re-inserted carrier in both proper frequency and
phase, whereas SSB tolerates some frequency offset without any
phase constraints.

-- 

-TV

Les Cargill  <lcargill99@comcast.com> wrote:

>That's my understanding - SSB was vaunted back in the CB radio days
>to have better range, which I assumed was due to better noise immunity.

It has better range than AM, other things being equal, because you 
are not wasting power transmitting a carrier with no information.

Double sideband suppessed carrier should be as good as SSB --
twice the noise bandwidth but a 2x redundant signal -- but you
get half as many users in a band.

If you can tolerate the audio squawkiness SSB is better, thus
is used by hams, CB'ers etc.

Steve

<danielot@gmail.com> wrote:

>I have the impression the Fig. 5 of the AARL paper you shared is still
>describing a baseband Hilbert transformer, which had to be implemented
>as a BPF due to physical limitations. 

Based solely on the diagram it's ambiguous I think.  It would be 
considered a "Low IF" (as opposed to "zero IF") receiver architecture 
which is the main design distinction, since there is only one ADC 
digitizing an IF signal.  But then there's some clever decimation.
Maybe the Hilbert transformer is at baseband ... but it does
still work having the Hilbert transformer at IF.

Steve

Tauno Voipio wrote:
> On 11.5.17 01:03, Steve Pope wrote:
>> <danielot@gmail.com> wrote:
>>
>>> Following DSPRelated moderator's advice, I'm moving this thread to the
>>> DSPrelated forum:
>>
>>> https://www.dsprelated.com/thread/2946/matlab-code-for-ssb-demodulation-using-phasing-method
>>>
>>>
>>> I've put there a much easier to follow code.
>>
>> Note that, not everybody on this newsgroup follows DSPRelated and
>> so by moving the thread, you have lost some of your potential
>> audience that might answer your questions.
>>
>> That said, have you tried using Mathwork's "fdatool" to create
>> the Hilbert transformer?  It should be dead easy to create
>> an SSB demodulator by this method.
>>
>> Steve
>
>
> I just wonder if the OP is barking up the wrong tree. There
> must be good grounds to suspect that there is noise/interference
> on one sideband but not on the other, and this is the situation
> where SSB demodulation will work.
>

That's my understanding - SSB was vaunted back in the CB radio days
to have better range, which I assumed was due to better noise immunity.

It's used in other systems for other reasons as well.

> The crud near the carrier could be from low modulation index
> PM sidebands, which look pretty much similar to AM sidebands,
> except for 90 degree phase difference compared to carrier.
> Noisy oscillators create easily small jitter for PM sidebands.
>

Wouldn't that in essence just be distortion[1] in the audio band,
due to errors in demodulation? It's roughly the same thing in
digital radio - although it's expressed there as higher BER
because those are mostly decoding in the phase domain.

[1] these guys call it noise, although I'd somehow suspect it'd
be correlated with the input signal, somehow:

http://www.robkalmeijer.nl/techniek/electronica/radiotechniek/hambladen/qst/1988/03/page14/

Sometimes people call distortion noise, which is understandable.

> ---
>
> And - yes, I'm not going to another forum.
>

-- 
Les Cargill

On Thursday, May 11, 2017 at 3:10:40 PM UTC-3, Evgeny Filatov wrote:
> Do you mean to say that you are measuring, say, X and Y displacements of 
> the beam (and that's why you need two channels), and then convert that 
> to polar coordinates?

(...)
 
> > If this explanation looks confusing, I'll try to put it in a more clear way in a longer text.
> 
> Definitely, you'd get better help here if you provide some more detailed 
> explanations.


On Thursday, May 11, 2017 at 4:37:24 PM UTC-3, Tauno Voipio wrote:
> 
> Is there any physical reason to think that some of the undesired
> signals are on one sideband only?


I'll try to formulate my issue better, but I'll need some time to write. I owe some explanations to you guys :)

Thanks for helping.

On Thursday, May 11, 2017 at 3:19:51 PM UTC-3, Steve Pope wrote:
> 
> Yes, if you have the Hilbert transformer at baseband, then you have
> additional difficulties in recovering components near DC, relative
> to having the Hilbert transformer at IF (as described in my last pair 
> of posts).

I finally made my code working! But then I realized it only works for baseband. I'm just wondering now how would it be possible to do the Hilbert transformer at IF as you said.

I have the impression the Fig. 5 of the AARL paper you shared is still describing a baseband Hilbert transformer, which had to be implemented as a BPF due to physical limitations. I think it would be impossible to design a LPF with +90 deg at DC+ frequencies and -90 deg at DC- frequencies.

Does it look right?

> 
> Generally -- if you're actually building a receiver -- you need
> an architecture and a frequency plan, and therefore face a decision
> between direct conversion and low-IF, ahead of your ADC.  For the
> SSB case these two choices correspond to the two possible placements
> for the Hilbert transformer described above.
> 
> If you're just doing a simulation for a proof of concept, then it may
> not matter yet, and you can simulate it either way.  But if you're
> pretty sure which way the final design will go, then save yourself
> some work and go in that direction now.

Thanks for the advice.

We have designed a receiver. By undersampling our 500 MHz carrier at ~221 MHz sampling rate (RF/864*383) we get the IF at 56.7 MHz (RF/864*98)). Then we downconvert to baseband digitally in a FPGA using standard AM demodulation using complex NCO. That's the point where I wanted to improve and use SSB demod. 

This is the project: http://www.ohwr.org/projects/bpm/wiki

Daniel

Tauno Voipio  <tauno.voipio@notused.fi.invalid> wrote:

>On 11.5.17 19:30, Steve Pope wrote:

>> This (if correct) is the "phasing method at IF" which was once popular
>> since the passband of the Hilbert transformer is narrow relative to
>> its operating frequency.  It was - back then - more difficult to
>> construct a Hilbert transformer at baseband.

>The essential main problem of Hilbert transform on baseband is that
>the minimum delay is a quarter wave at the lowest frequency of
>interest, because otherwise we'll need a non-causal filter.

Good point.  Although (thought experiment) if you mix it back
up to RF, have an equivalent Hilbert transformer with less
latency, and then mix it back down to baseband, do you not
then have a causal version of this "non causal" filter?

Heh heh, that would be a good exam question.

Steve