Direct Conversion Receiver

On Thu, 8 Nov 2012 11:09:05 -0600, "Vladimir Vassilevsky"
<nospam@nowhere.com> wrote:

>
>"Eric Jacobsen" <eric.jacobsen@ieee.org> wrote in message 
>news:509bdd0a.313794014@www.eternal-september.org...
>> On Thu, 8 Nov 2012 08:54:52 -0600, "Vladimir Vassilevsky"
>> <nospam@nowhere.com> wrote:
>>
>>>
>>><kalvin.news@gmail.com> wrote in message
>>>news:fdc57980-fc10-460a-94be-ae756da584c2@googlegroups.com...
>>>> Nice document about software defined radios and direct-conversion
>>>> receivers:
>>>>
>>>> "A Software-Defined Radio for the Masses, Part 1 - ARRL"
>>>> http://www.arrl.org/files/file/Technology/tis/info/pdf/020708qex013.pdf
>>>>
>>>> Especially Tayloe-detector is very simple and straightforward
>>>> mixer/detector concept (Fig. 10 and Fig. 14).
>>>
>>>It is a strange idea trying to receive anything with crappy direct
>>>conversion receiver positioned close to PC.
>>>
>> So use a good one, instead.
>
>Indeed. It took us a long time to find a 802.11 solution that would be good, 
>reliable and performing to the standard.

In a PC?  Teenagers get these things working all the time.   I've
never had trouble with one since 802.11g came out.

What part of the standard is failing for you?  Which standard?

>>802.11 cards are universally direct
>> conversion, and they work fine in PCs and laptops and lots of other
>> similar equipment.
>
>That's why there are so many poorly performing cards that hiccup at tabletop 
>distances.

If you buy cheap crap, expect poor performance.  If you don't know the
limitations or how to apply it, don't expect quiet free space
performance.

>> It just has to be engineered competently.
>
>So, where is it? Could you please engineer one for me.

Don't need to.   There are a lot of good products on the market that
have been through subtantial verification testing.


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

A "true" direct conversion (not "RF sampling") is great if you order 1M
units. Let's do an ASIC and put a tuneable analog filter on-chip. No need
for an IF SAW filter and some other stuff, you save maybe a dollar per
unit. Big bucks. But we'll need a few dozen guys to tie up all the loose
ends, and expect a respin or two. The devil is in the details. 

Now if I want to order just one of them, it'll still save me one dollar per
unit. One cup of coffee at McDonald's, best enjoyed after a direct
conversion receiver death march project.
 
No, I doubt anybody would propose DC without very good reasons, such as
making a high-volume product. Definitely not for performance, if I could
use a superhet instead.

The WLAN standard -is- meant for high volume products, here DC is suddenly
essential for market acceptance. Single chip CMOS, integrated PA, the
standard goes out of its way so that you can do it, with an on-chip
oscillator that would be traded as 'narrow-band noise source' in the
cellular community (if all you know is GSM...)

Knowing no details, just to build "a" receiver, my strategy would be: Find
a ready-made superhet frontend or build one from the minicircuits
catalogue. Find a suitable off-the-shelf oscillator and IF filter that
comes close enough, sample at IF and spend the $$$ on an ADC that has more
effective bits instead of a higher rate. Then, use the added dynamic range
for digital AGC.

On Thu, 08 Nov 2012 12:49:23 -0600, "mnentwig" <24789@dsprelated>
wrote:

>A "true" direct conversion (not "RF sampling") is great if you order 1M
>units. Let's do an ASIC and put a tuneable analog filter on-chip. No need
>for an IF SAW filter and some other stuff, you save maybe a dollar per
>unit. Big bucks. But we'll need a few dozen guys to tie up all the loose
>ends, and expect a respin or two. The devil is in the details. 
>
>Now if I want to order just one of them, it'll still save me one dollar per
>unit. One cup of coffee at McDonald's, best enjoyed after a direct
>conversion receiver death march project.
> 
>No, I doubt anybody would propose DC without very good reasons, such as
>making a high-volume product. Definitely not for performance, if I could
>use a superhet instead.
>
>The WLAN standard -is- meant for high volume products, here DC is suddenly
>essential for market acceptance. Single chip CMOS, integrated PA, the
>standard goes out of its way so that you can do it, with an on-chip
>oscillator that would be traded as 'narrow-band noise source' in the
>cellular community (if all you know is GSM...)
>
>Knowing no details, just to build "a" receiver, my strategy would be: Find
>a ready-made superhet frontend or build one from the minicircuits
>catalogue. Find a suitable off-the-shelf oscillator and IF filter that
>comes close enough, sample at IF and spend the $$$ on an ADC that has more
>effective bits instead of a higher rate. Then, use the added dynamic range
>for digital AGC.

Like most technologies, direct conversion has its tradeoffs, as you
point out.   High-volume, low-cost products being a prime example of
where they can make the most sense.   Applications that need a wide
tuning range are another, as that's an area where it's harder to make
a superhet work than a direct-conversion system.

As you mention, the WLAN standards were developed with direct
conversion in mind.  I don't know why someone would think that the
implementations don't meet the standards.


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

On 11/7/2012 2:04 PM, Randy Yates wrote:
> "mnentwig"<24789@dsprelated>  writes:
>
> Hi Markus,
>
>>>> anti-aliasing requirements
>>
>> that would be the least of my concerns, run a fixed-frequency NCO to shift
>> the whole band near 0 Hz still at the high rate, then lowpass -/ decimate
>> in multiple stages (that said, I've never actually implemented this, and
>> there may be better ways)
>
> Well, my idea wouldn't require an NCO or a mixer at all. By careful
> choice of the high sample rate, one could simply perform a complex,
> one-sided bandpass filter (which results in a complex output) at the
> high sample rate followed by decimation.
>
> Yes (Dale), this would require a complex filter in general, but remember
> the "effective" (i.e., computational) length of the filter is also
> decimated by the decimation factor if you use polyphase filtering. One
> may be able to do get away with doing this filter in stages as well,
> depending on the frequency planning. Don't know - haven't thought
> that all the way through yet.
>
> Either way, an FPGA should be able to do the job.

I'm not sure how this is significantly different from an NCO except that 
it would require instead a VCO to achieve your variable sample rate. 
Also, the filter would need to be *very* effective to exclude all the 
aliased energy that would be picked up in the same bandwidth of all the 
folding of the spectrum resulting from the under-sampling.  Yes, taking 
advantage of polyphase filtering would help, but this would be one hell 
of a filter given the disparity of the bandwidth and the sample rate. 
By doing the decimation in stages it might become practical.


>> no, I think getting the RF signal through the ADC is the hard part, unless
>> the radio requirements are very easy.
>> The main difficulty in receiver design is that I have to be prepared for
>> any likely combination of unknown interferers.
>
> I was also planning on doing the preselector in stages based on
> oversampling the desired RF bandwidth by a factor of two. (Isn't modern
> technology wonderful that we can do this type of thing these days?!)
> First stage would be the analog that would have the desired stopband
> attenuation at 2x the highest frequency. That should afford a) a more
> reasonable design, or b) higher stopband attenuation depending on your
> goals. The second stage would be "digital preselector" after the ADC,
> complex, one-sided bandpass filtering the desired RF bandwidth further.
> The decimation step would purposely bandpass alias the desired RF band
> to baseband.

Yes, what you describe sounds good, but the interferer issue remains. 
The classic problem is two signals closely spaced in frequency.  The one 
you don't want to hear is very strong and the one you want to hear is 
weak.  My FM radio station is 50 miles away and right next on the dial 
(+200 kHz) to some loudmouth screaming opinion crap from a mile away.  I 
want to hear my music.  His opinion crap overloads the ADC or if the RF 
gain is reduced my signal isn't strong enough to hear.

Rick

On 11/6/2012 2:20 PM, Randy Yates wrote:
> "mnentwig"<24789@dsprelated>  writes:
>
>> Hi,
>>
>> I did a quick calculation: If I'd plug it straight to a terminated 50 ohms
>> antenna, I'd get fullscale with a ~3 dBm CW signal, and a noise figure of
>> 48 dB.
>> Put 20 dB of gain in front of it and it clips at -17 dBm with 28 dB noise
>> figure. Which is still 20 dB too high for any self-respecting receiver, but
>> this depends on the application. The first number, tolerance to blockers,
>> may be already too bad when radio, TV, WLAN or cell phone towers are
>> near.
>
> Perhaps I was too tunnel-visioned when I wrote my original query. I
> didn't mean just stick the antenna into the ADC.
>
> I presumed you would have some sort of "sufficient" preselector filter.
> I also presumed one would have a front-end gain element that established
> system noise figure (mainly) and enough gain to overcome the insertion
> loss, and then quite a bit, of the preselector. Then possibly a signal
> conditioning amp prior to the ADC (after the preselector).
>
>> Besides noise, linearity may be a problem - the DAC isn't perfect -
>
> Did you mean ADC?
>
>> and the preamplifier makes it even worse. With a wide-band frontend,
>> there is a huge number of input tone combinations (adjacent channels,
>> TV, whatever) that can cause inter- or crossmodulation products.
>
> Isn't that true for either DC or superhet topology? One needs a preamp
> either way.
>
>> Your clock and timing uncertainty need to be as clean as a local oscillator
>> in a conventional receiver. This alone often rules out the idea.
>
> Why? Why not make the ADC clock as clean as an LO?

If the LO is off by 1 Hz, the resulting signal is off by 1 Hz.  If your 
sample clock is off by 1 Hz, how much off is your signal after 
decimation by... what decimation factor were you going to use?  I 
believe I read in another post that you were going to oversample by 2 (I 
assume this means your signal will initially be at Fs/4), then decimate 
to DC.  That isn't enough to tell me the decimation.  To figure that we 
would need to know the ratio of the initial and final bandwidths.


>> I suspect you'll be better off using ready-made RF frontend components and
>> then digitize at some IF after filtering.
>
> Well the devil is in these details, I suspect.

For sure!  Greatly depends on the frequency of interest and bandwidth.

Rick

On Thu, 8 Nov 2012 08:54:52 -0600, "Vladimir Vassilevsky"
<nospam@nowhere.com> wrote:

>> Especially Tayloe-detector is very simple and straightforward 
>> mixer/detector concept (Fig. 10 and Fig. 14).
>
>It is a strange idea trying to receive anything with crappy direct 
>conversion receiver positioned close to PC.
>
>VLV
>
In practice it does not seem a problem. The 7mhz version (Soft Rock) I
have is mounted in the PC and powered from the PC. 
John Ferrell W8CCW

"mnentwig" <24789@dsprelated> writes:
> [...]
> Definitely not for performance, if I could use a superhet instead.

Assuming you have enough dynamic range for the near/far problem in a DC
receiver, how does a superhet give you better performance?

We seem to be ignoring the image frequency problem here. Even with a
properly-designed frequency plan, where the image band is outside the
preselector band, the preselector isn't going to be perfect. Thus a
_hit_ in performance for the superhet approach. You also have the
problem of controlling third-order intercept spurs in the mixer.
-- 
Randy Yates
Digital Signal Labs
http://www.digitalsignallabs.com

>> how does a superhet give you better performance?

for example 
* DC offset / flicker noise / LO leakage / highpass frequency response
(once you track and / or cancel DC) 
  DC-friendly radio standards usually don't put a subcarrier on top of the
channel frequency, for those reasons
* near-infinite fractional bandwidth (it is probably easier to get your
gain on a narrow fractional bandwidth at IF than from DC to x MHz).
* 2nd order distortion (envelope squaring). There are multiple mechanisms
that interact, and sorting through them has been the topic of more than one
doctoral thesis.
* IQ imbalance
That's one of the few advantages on ASICs over discrete components that's
in favour of DC: ASIC component tolerances are all over the place, but you
can match identical components with very high accuracy.
* phase noise (the LO with most of the PN around it sits inside the wanted
signal. In a superhet, the LO is one IF spacing distant from the signal)
* need for quadrature LO. Again, ASICs are good at this (use higher
frequency Sx and digital divider)
* the list isn't complete, just what came to mind immediately.

BTW, any spectrum analyzer I know - example for not-so-high volume,
discrete high performance - uses a superhet frontend, usually with multiple
IFs.

There are a lot of amateur radio circuits and kits out there to experiment with. I think a combination of both theoretical and practical knowledge will give the best outcome.