From: Vladimir Vassilevsky on


Paul Keinanen wrote:
> On Tue, 13 Jul 2010 15:44:19 -0500, Vladimir Vassilevsky
> <nospam(a)nowhere.com> wrote:
>
>
>>For ~10 bit accuracy, the PWM rate must be ~20 times higher then the
>>highest frequency of the signal.
>>
>>Generating 10-bit linear ramp at 2 GHz is nontrivial.
>
>
> Why that high ?

If the PWM is demodulated with LPF, then it is inherently nonlinear.
Higher clock frequency - better linearity.

> The input signal needs to be band limited to 150 MHz, then sampled
> (but not quantized) at something above 300 MHz (Nyquist), then feed it
> to a comparator with a 300+ MHz linear ramp and the output drives the
> LD.
>
> At receiver end, a sample&hold circuit is also required and a LPF.

You can generate synchronized ramp voltages at Tx and Rx sides of the
link. Those ramps have to be identical, but they don't have to be
linear. On Tx side, you compare the ramp to the sample-holded input
analog signal.On the Rx side, you latch the ramp voltage into sample
hold by the edge of the incoming signal. This produces nonuniformly
sampled signal which has to be converted to uniform sampling by
integrate and dump circuit, and then restored to analog by means of LPF.
This scheme doesn't produce any inherent distortion but from
imperfection of timing, sample holds and integrate and dumps. However
achieving that in practice at 300+ MHz could be problematic.


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com



From: Paul Keinanen on
On Wed, 14 Jul 2010 09:19:12 -0500, Vladimir Vassilevsky
<nospam(a)nowhere.com> wrote:

>
>
>Paul Keinanen wrote:
>> On Tue, 13 Jul 2010 15:44:19 -0500, Vladimir Vassilevsky
>> <nospam(a)nowhere.com> wrote:
>>
>>
>>>For ~10 bit accuracy, the PWM rate must be ~20 times higher then the
>>>highest frequency of the signal.
>>>
>>>Generating 10-bit linear ramp at 2 GHz is nontrivial.
>>
>>
>> Why that high ?
>
>If the PWM is demodulated with LPF, then it is inherently nonlinear.
>Higher clock frequency - better linearity.

The received analog PWM can be converted to analog pulse amplitude
modulation (PAM) with theoretically any value (not just some few
discrete quantized values). Those PAM pulses are copies of those
pulses at the sampler output at the Tx end (with some inaccuracy).

Gating those very short PAM pulses (with the rest of the pulse cycle
time as zero) into an analog LPF will restore the original bandlimited
signal at the Tx end.

Thus no oversampling required.

From: Vladimir Vassilevsky on


Paul Keinanen wrote:

> On Wed, 14 Jul 2010 09:19:12 -0500, Vladimir Vassilevsky
> <nospam(a)nowhere.com> wrote:
>
>
>>
>>Paul Keinanen wrote:
>>
>>>On Tue, 13 Jul 2010 15:44:19 -0500, Vladimir Vassilevsky
>>><nospam(a)nowhere.com> wrote:
>>>
>>>
>>>
>>>>For ~10 bit accuracy, the PWM rate must be ~20 times higher then the
>>>>highest frequency of the signal.
>>>>
>>>>Generating 10-bit linear ramp at 2 GHz is nontrivial.
>>>
>>>
>>>Why that high ?
>>
>>If the PWM is demodulated with LPF, then it is inherently nonlinear.
>>Higher clock frequency - better linearity.
>
>
> The received analog PWM can be converted to analog pulse amplitude
> modulation (PAM) with theoretically any value (not just some few
> discrete quantized values). Those PAM pulses are copies of those
> pulses at the sampler output at the Tx end (with some inaccuracy).

Yes, of course. PWM could be perfectly converted to/from regular PAM by
means of sample holds and/or integrate and dumps. As you noted, the
keyword is "accuracy". I am not sure what is feasible at 300+ MHz.

> Gating those very short PAM pulses (with the rest of the pulse cycle
> time as zero) into an analog LPF will restore the original bandlimited
> signal at the Tx end.
>
> Thus no oversampling required.

VLV
From: Joerg on
John Larkin wrote:
> On Tue, 13 Jul 2010 13:31:43 -0700, Tim Wescott <tim(a)seemywebsite.com>
> wrote:
>
>> On 07/13/2010 08:59 AM, John Larkin wrote:
>>> On Tue, 13 Jul 2010 08:43:13 -0700, Tim Wescott<tim(a)seemywebsite.com>
>>> wrote:

[...]

>>>> I think the most important "high-end" math tool in this case is the
>>>> consultant's brain -- Scilab you can get off the web for free and go buy
>>>> a nice car with the money you would have spent to buy Matlab. With
>>>> either Scilab or Matlab you still need a nice squishy pile of neurons
>>>> that knows how to feed in the questions the right way and interpret the
>>>> results.
>>> Hence the option to have a consultant, a real RF signals guy familiar
>>> with the tools, furnish the neurons.
>> It all sounds interesting. I assume (well, gather from other posts in
>> the thread) the ultimate goal is to be able to send nice clean analog
>> signals while dodging any need to digitize?
>
> Preferebly without digitizing. At our bare-minimum 100 MHz signal
> bandwidth, and 10 bits, sampling at 250 MHz, just past Nyquist, we'd
> be shipping 2.5 gbits/second. That's feasible, but I thought we'd
> consider FM too, which would be simpler to do.
>

Why not PWM?

[...]

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.
From: Joerg on
Vladimir Vassilevsky wrote:
>
>
> Joerg wrote:
>
>> Vladimir Vassilevsky wrote:
>>> Joerg wrote:
>>>> Vladimir Vassilevsky wrote:
>>>>
>>>>> For ~10 bit accuracy, the PWM rate must be ~20 times higher then the
>>>>> highest frequency of the signal.
>>>>>
>>>> Why that high? I bet John would spring for more than an RC lowpass
>>>> here :-)
>>>>
>>>
>>> Even if he puts ideal brickwall, that won't help.
>> Of course not, but the rate doesn't have to be 20 times higher.
>> Again, how do you arrive at 20 times?
>
> PWM is a kind of angular modulation. As such, it creates infinite
> sidebands on both sides of the carrier. Some part of the lower sideband
> inevitably falls into the bandwidth of the useful signal. How much of
> trash gets into the signal? It depends. Ballpark: for 60dB of rejection,
> the PWM carrier should be ~ x20 times of the highest signal frequency.
>

I've done a _lot_ better than 60dB. Ok, those ones weren't over fiber
but transformers and coax. The transformers were the limiting factor but
it simply was not necessary to use a 20x carrier.


> >>>>(This accounts for negative feedback in transmitter. Without NFB, the
>>>>> results are going to be several times worse).
>>>>
>>>> Well, yeah, it would have to be some, as John called it, one-shot on
>>>> steroids. This will not be very accurate and linear so it needs to be
>>>> servoed at the transmitter.
>>>
>>> Loop cutoff ~ PWM rate/8. Do the math.
>>>
>> This part has to be a little more nifty than just a simple loop. Ideally
>> there should be a characteriziation plus tempco-LUT so you can "steer"
>> the PWM generator to the points where the signal wants it to, with
>> little residual error.
>
> This could buy some accuracy. But, still loop cutoff ~ PWM rate/8. Or,
> if you want to really push phase margin, PWM rate/6. Consequently, the
> feedback is going to be quite shallow and it can't be the cure for all
> sins of PWM.
>

For simple loops, yes. But even if the PWM was a GHz that ain't
considered impossible these days. You'd have to make the controlled
oscillator out of inverters though. Maybe even accept a non-constant
frequency. The FCC patrol isn't out there on those fibers :-)

[...]

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.