I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal
rather than mixing it with IF as is done in a superhet, then use DSP
techniques to extract the audio from the signal (don't know if one
would digitize the original signal itself, or to somehow subtract the
carrier before digitizing.)

I have no idea if it is possible to accurately digitize such low
voltage signals (at least in a practical sense). But if doable, I
surmise that clever design may meet the various requirements of
good sensitivty, selectivity, dynamic range and audio fidelity.

Again, it is patently obvious that I border on clueless on this
topic, so feel more than free to set me straight.

Eric S.

Eric S. wrote:
I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal
rather than mixing it with IF as is done in a superhet, then use DSP
techniques to extract the audio from the signal (don't know if one
would digitize the original signal itself, or to somehow subtract the
carrier before digitizing.)

I have no idea if it is possible to accurately digitize such low
voltage signals (at least in a practical sense). But if doable, I
surmise that clever design may meet the various requirements of
good sensitivty, selectivity, dynamic range and audio fidelity.

Again, it is patently obvious that I border on clueless on this
topic, so feel more than free to set me straight.

Eric S.

Hi Eric,

Well, I think you have a clue, but not much beyond that. ;-)

In the spirit of continuing mental exploration of your ideas, here is
what you'd have to do to build a fully digital AM radio:

There are essentially two operations that have to occur: (Actually
there's a third one; amplification. But I'll leave that one alone,
assuming that we'll be using a "suitable" broad-band preamplifier of
some sort.)

1) Tuning
2) Detection

The first operation separates out the one station you want to listen to,
The whole idea of the superheterodyne is merely to make this operation
easier, in that the main amplification only has to be done at a single
frequency rather than having to track multiple amplifier stages as in
the TRF (tuned RF) design. The second operation demodulates it so you
can hear it.

It's conceivable to do the tuning operation using DSP. But the sample
rate would have to be very high - at least twice the maximum frequency,
which translates to over 3 MHz. (As a point of comparison, hi-fi audio
is now done at 96 kHz, i.e. you'd have to sample over 30 times as fast!)

Then you'd need a software bandpass filter using FFT to select the
frequency range corresponding to your station of interest.

OK, now you've got the modulated RF. You have to demodulate it. This is
not a matter of subtracting out the RF, since the RF and AF are *mixed*,
not *added*. Rather, you'd have to simulate the detector in a
conventional set, by introducing a non-linearity. The simplest would be
to "chop off" the top or bottom of the waveform, and apply a low-pass
filter (again using FFT software).

Finally you'd have to restore the DC level, which can again be
accomplished in software.

But -- can this all be done in real time? I have my doubts. Perhaps with
a big Linux cluster or other distributed computing system you could
approach real-time processing at this speed, but even that would be a
chore. (I can see it now... 100 X-boxes linked together to digitally
simulate an AA5.)

Rather like using a crane to pick up a pearl. Possible, but fiddly.

Cheers,
Fred
--
+--------------------------------------------+
| Music: http://www3.telus.net/dogstarmusic/ |
| Projects: http://dogstar.dantimax.dk |
+--------------------------------------------+

The Symphony digital radio chip set from Motorola already does this,
although it uses a superhetrodyne front end to convert to a high frequency
IF, maybe 10.7 MHz I don't remember exactly, before sampling. Depending
on how the A/D converter in this chip set works, and which chip it is on,
it might be possible to adapt the A/D, DSP, and D/A portions of this chip
set to work with your TRF tube front end idea.

There are also other tube/DSP architecture's you might want to consider.


Regards,

John Byrns


In article ,
wrote:

I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal
rather than mixing it with IF as is done in a superhet, then use DSP
techniques to extract the audio from the signal (don't know if one
would digitize the original signal itself, or to somehow subtract the
carrier before digitizing.)

I have no idea if it is possible to accurately digitize such low
voltage signals (at least in a practical sense). But if doable, I
surmise that clever design may meet the various requirements of
good sensitivty, selectivity, dynamic range and audio fidelity.

Again, it is patently obvious that I border on clueless on this
topic, so feel more than free to set me straight.

Eric S.


Surf my web pages at, http://users.rcn.com/jbyrns/

http://www.eetimes.com/semi/news/OEG20020930S0034

"Eric S." wrote in message
...
I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal

If you mean a tuned-RF design, then there's not much to do after the tube
TRF stages but "detect" the signal, which only requires a diode. Digitizing
the RF signal, just to detrect it, is a bit of overkill.

If you mean jsut having a tube RF amplifier stage, then go digital, you're
going to have some rather difficult challenges:

#1: You've got to digitize the RF at at least twice the carrier frequency,
so say 3.2M samples/sec.
#2: The RF signal after just one tuned stage is going to consist of maybe
100KC wide spectrum, if there's a weak station and a strong station in
there, you could easily have a 50 to 60 db amplitide range, so you'll need a
mighty accurate and many bit wide A/D converter-- 12 bits at least, 16
desireable. These have come down a lot in price, but they're still mighty
sophisticated and cranky beasts.

You then have to process these samples. Not impossible, for a really good
DSP or Pentium. Programming a good sharp IF filter is going to chew up most
of the CPU power, but it just might be doable.

But where's the fun in all this bit-twiddling?

Maybe if you do all Analog stuff by day, a digital hobby might be bearable.
But some of us work all day with bits-- we'd rather get away from
programmable things at home!

Regards,


George

Eric S. wrote:

I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal
rather than mixing it with IF as is done in a superhet, then use DSP
techniques to extract the audio from the signal (don't know if one
would digitize the original signal itself, or to somehow subtract the
carrier before digitizing.)

I have no idea if it is possible to accurately digitize such low
voltage signals (at least in a practical sense). But if doable, I
surmise that clever design may meet the various requirements of
good sensitivty, selectivity, dynamic range and audio fidelity.



You'd need a ADC capable of sampling at around 1.8MHz clock speed, and you'd
need it to have a dynamic range of at least 24 bits. (24 bits because
you'd want to
have decent digitalization of weak signals and still cope with strong
signals).
Actually, if you drive the ADC with the usual superhet local oscillator,
the sample
rate would Nyquest alias with the station carrier to yield a digital
"IF" at 455KHz.
And then you could do DSP on that. Having the "digital IF" at a fixed
frequency
would make the DSP algorythms easier and better to write. If we did
this, we
might as well do the mixer using analog circuits and then digitize the
analog IF.

In any event, you'd need either a huge dynamic range (lots of bits) or
analog
front end and analog IF filtering to eliminate strong local stations
near on the
dial to some interesting DX. Or else the local station will saturate
the ADC
and then you'd get very choppy reception.

"Eric S." wrote:

I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal
rather than mixing it with IF as is done in a superhet, then use DSP
techniques to extract the audio from the signal (don't know if one
would digitize the original signal itself, or to somehow subtract the
carrier before digitizing.)

I have no idea if it is possible to accurately digitize such low
voltage signals (at least in a practical sense). But if doable, I
surmise that clever design may meet the various requirements of
good sensitivty, selectivity, dynamic range and audio fidelity.

Again, it is patently obvious that I border on clueless on this
topic, so feel more than free to set me straight.

Eric S.

The idea of digitising ALL of the signal received by an antenna
and then picking out which one of the thousands present isn't
new, using digital processing.
The process should not upset the amplitude level changes to the
recieved signal, or the audio recovered will be compressed.

Doing the digital processing with tubes is a waste of glassware
because the active devices work as switches.
A chip AD converter is better....
There are already cards which plug into a PC, which allow a virtual radio

to appear on the screen, and the mouse is used to tune the "set".

But to make a decent AM radio using tubes requires good selectivity
and a an audio BW of 10 kHz, which will allow good reception of local
stations only
which are say 50 kHz apart.
It would be more effective to simply use superhet
with a twin tuned input coils, to get a broad tuning BW, say 20 kHz.
The next stage can be a twin triode cathode coupled RC stage,
but with AVC applied, one of the vari-mu twin triodes are fine.

The next stage can be a 6BE6 or 6AN7 mixer with fixed bias, then have
perhaps an IF of 2 MHz, which would give a BW of at least 15 kHz each.
The IF tube is best being a pentode like the 6AU6 or 6BX6 using
unbypassed cathode bias,
and agian, no AVC applied, but applied only to the input triode stage.

I have such a radio I built from scratch, but I stuck with
455 kHz IFTs, but IFT1 has a sliding secondary coil to vary the
BW.
Anyway, this radio also has a low distortion AM detector
by feeding the IF amp output to a cathode follower, then a germanium
diode
and RC network.
The AF has an SET EL34, driven by 12AX7, some feedback,
and a bass reflex speaker with a proper tweeter.
The sound is very similar to FM radio in clarity, when
clear signals are transmitted, which isn't all that often.
The sound is FAR better than all the crummy SS based AM sets I have ever
used,
and better than all the AM tubed radios I have listened to.
It even beats a Quad AM tuner, which I have to say isn't too bad.

Some folks believe in TRF, but you'd need 5 tuned circuits to
compete with a superhet's critically tuned IFTs, and getting wide AF BW
is somewhat difficult, since making all those variable tuning gangs track
properly
is a PITA.

Patrick Turner.

"Eric S." wrote in message
...

I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.


** Has anyone ever mentioned the Aussie built Audiosound "AM100" tube
AM tuner ?? Designed for local conditions where AM is of wide audio
bandwidth and not crowded with interfering signals.

The design uses just three tubes, a 6N8, a 6AN7 and an EM84 tuning
indicator. The PSU uses diodes and the detector is a biased germanium diode
operating at high level to reduce THD to under 1% at full modulation.

There is a balanced loop antenna which dramatically reduces atmospheric
and most man made noises plus a passive ( LC type ) notch filter to sharply
knock out 9 kHz and above for night time listening.

IME long distance AM is possible but in no way is it hi-fi quality -
however with a large enough loop it is just possible to hear signals from
New Zealand in Sydney on the set.




......... Phil

"George R. Gonzalez" writes:


"Eric S." wrote in message
.. .
I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal

snip

Tube/digital would be lots of fun, but use the best of both. SuperHet
down to base band and then use DSP to select AM/FM/CW, etc. with
programmable filters. Run the RF gains with lots of local feedback and
low gain to get low intermod. Use a DSP controlled analog preselector.
Use a digital LO and micro control of the tracking.

Lots of work, but you can most of the algorithms in "digital radio" books
and IEEE articles. The tube stuff is well known.

Have fun.

Steve.

--
Steven D. Swift, , http://www.novatech-instr.com
NOVATECH INSTRUMENTS, INC. P.O. Box 55997
206.301.8986, fax 206.363.4367 Seattle, Washington 98155 USA

Fred Nachbaur wrote:

Eric S. wrote:
I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal
rather than mixing it with IF as is done in a superhet, then use DSP
techniques to extract the audio from the signal (don't know if one
would digitize the original signal itself, or to somehow subtract the
carrier before digitizing.)

I have no idea if it is possible to accurately digitize such low
voltage signals (at least in a practical sense). But if doable, I
surmise that clever design may meet the various requirements of
good sensitivty, selectivity, dynamic range and audio fidelity.

Again, it is patently obvious that I border on clueless on this
topic, so feel more than free to set me straight.

Eric S.

Hi Eric,

Well, I think you have a clue, but not much beyond that. ;-)

In the spirit of continuing mental exploration of your ideas, here is
what you'd have to do to build a fully digital AM radio:

There are essentially two operations that have to occur: (Actually
there's a third one; amplification. But I'll leave that one alone,
assuming that we'll be using a "suitable" broad-band preamplifier of
some sort.)

1) Tuning
2) Detection

The first operation separates out the one station you want to listen to,
The whole idea of the superheterodyne is merely to make this operation
easier, in that the main amplification only has to be done at a single
frequency rather than having to track multiple amplifier stages as in
the TRF (tuned RF) design. The second operation demodulates it so you
can hear it.

It's conceivable to do the tuning operation using DSP. But the sample
rate would have to be very high - at least twice the maximum frequency,
which translates to over 3 MHz. (As a point of comparison, hi-fi audio
is now done at 96 kHz, i.e. you'd have to sample over 30 times as fast!)

Then you'd need a software bandpass filter using FFT to select the
frequency range corresponding to your station of interest.

OK, now you've got the modulated RF. You have to demodulate it. This is
not a matter of subtracting out the RF, since the RF and AF are *mixed*,
not *added*. Rather, you'd have to simulate the detector in a
conventional set, by introducing a non-linearity. The simplest would be
to "chop off" the top or bottom of the waveform, and apply a low-pass
filter (again using FFT software).

Finally you'd have to restore the DC level, which can again be
accomplished in software.

But -- can this all be done in real time? I have my doubts. Perhaps with
a big Linux cluster or other distributed computing system you could
approach real-time processing at this speed, but even that would be a
chore. (I can see it now... 100 X-boxes linked together to digitally
simulate an AA5.)

Rather like using a crane to pick up a pearl. Possible, but fiddly.

Cheers,
Fred
--

Fred, the Microdyne RCB-2000 telemetry receivers I worked on had a 90
MHz IF that fed a D/A card. then it used a pair of FIR filters to select
the desired signal. Another pair of FIR filters followed that to shape
the output to the demod to recover the data. The IF bandwidth was
software selectable from 10 KHz to 40 MHz, and the "Video" filters that
followed worked the same. With the FIR filters, a lot of the processing
load was removed from the microprocessors, which were 16 bit Motorola
chips. It took more processing power to handle the GUI and to drive the
front panel display. It used a Cyrix processor, and ran under embedded
NT. There were six processor chips in the radio on a custom buss,
controlled by a 19.2 Kb serial port on the embedded controller. All
this, for just $80,000.00 US

--


Michael A. Terrell
Central Florida

Michael A. Terrell wrote:

Fred Nachbaur wrote:

[...]

Hi Eric,

Well, I think you have a clue, but not much beyond that. ;-)

In the spirit of continuing mental exploration of your ideas, here is
what you'd have to do to build a fully digital AM radio:

There are essentially two operations that have to occur: (Actually
there's a third one; amplification. But I'll leave that one alone,
assuming that we'll be using a "suitable" broad-band preamplifier of
some sort.)

1) Tuning
2) Detection

The first operation separates out the one station you want to listen to,
The whole idea of the superheterodyne is merely to make this operation
easier, in that the main amplification only has to be done at a single
frequency rather than having to track multiple amplifier stages as in
the TRF (tuned RF) design. The second operation demodulates it so you
can hear it.

It's conceivable to do the tuning operation using DSP. But the sample
rate would have to be very high - at least twice the maximum frequency,
which translates to over 3 MHz. (As a point of comparison, hi-fi audio
is now done at 96 kHz, i.e. you'd have to sample over 30 times as fast!)

Then you'd need a software bandpass filter using FFT to select the
frequency range corresponding to your station of interest.

OK, now you've got the modulated RF. You have to demodulate it. This is
not a matter of subtracting out the RF, since the RF and AF are *mixed*,
not *added*. Rather, you'd have to simulate the detector in a
conventional set, by introducing a non-linearity. The simplest would be
to "chop off" the top or bottom of the waveform, and apply a low-pass
filter (again using FFT software).

Finally you'd have to restore the DC level, which can again be
accomplished in software.

But -- can this all be done in real time? I have my doubts. Perhaps with
a big Linux cluster or other distributed computing system you could
approach real-time processing at this speed, but even that would be a
chore. (I can see it now... 100 X-boxes linked together to digitally
simulate an AA5.)

Rather like using a crane to pick up a pearl. Possible, but fiddly.

Cheers,
Fred
--


Fred, the Microdyne RCB-2000 telemetry receivers I worked on had a 90
MHz IF that fed a D/A card. then it used a pair of FIR filters to select
the desired signal. Another pair of FIR filters followed that to shape
the output to the demod to recover the data. The IF bandwidth was
software selectable from 10 KHz to 40 MHz, and the "Video" filters that
followed worked the same. With the FIR filters, a lot of the processing
load was removed from the microprocessors, which were 16 bit Motorola
chips. It took more processing power to handle the GUI and to drive the
front panel display. It used a Cyrix processor, and ran under embedded
NT. There were six processor chips in the radio on a custom buss,
controlled by a 19.2 Kb serial port on the embedded controller. All
this, for just $80,000.00 US


Thanks, Michael (and everyone else). Seems I have indeed been "out of
the loop" too long, as regards the capabilities of present-day digital
technology. (Heck, I can remember when an 8 MHz clock was screaming fast!)

I also received a fascinating reply by email, I'll try to convince him
to post it here because it truly was an eye-opener for me.

Cheers,
Fred
--
+--------------------------------------------+
| Music: http://www3.telus.net/dogstarmusic/ |
| Projects: http://dogstar.dantimax.dk |
+--------------------------------------------+

On Sun, 05 Oct 2003 16:36:44 +0000, the highly esteemed Eric S.
enlightened us with these pearls of wisdom:

I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal
rather than mixing it with IF as is done in a superhet, then use DSP
techniques to extract the audio from the signal (don't know if one
would digitize the original signal itself, or to somehow subtract the
carrier before digitizing.)

I have no idea if it is possible to accurately digitize such low
voltage signals (at least in a practical sense). But if doable, I
surmise that clever design may meet the various requirements of
good sensitivty, selectivity, dynamic range and audio fidelity.

Again, it is patently obvious that I border on clueless on this
topic, so feel more than free to set me straight.

Eric S.

Conceptually, it is not a bad idea - a broadband RF amplifier for a front
end, rolled off below 400KHz and above 1.7MHz, whose signal is then
digitized. The digitized RF is sent through the DSP, which does your
bandpass filtering and detection in the digital domain, and outputs the
"detected" signal into a conventional 16/44.1KHz (or less if you want)
DAC.

The main problem isn't the RF front end or the DSP - the front end can be
made either with tubes, JFETs, bipolars, or easiest of all with some good
high-speed op-amps (especially since we are talking 2 MHz here), and
sufficently powerful, inexpensice DSPs are readily available from TI, AD,
Mot, etc. The problem is the ADC. You will need at least 3.5MSPS - not
too bad - with sufficent dynamic range, which IS a problem. 10 and 12 bit
converters are readily available that can digitize at that speed, but 12
bits isnt enough; the dynamic range is only a little over 70db. 16 bits
with a good SFDR is what is really needed, since the incoming signal
can have a very large variation is signal strength - you dont want the
strongest to exceed the ADCs input range, while you want the weaker
signals to toggle more than a couple bits. 18 bits would be even better,
but good luck finding one. In the 16 bit realm, you could use something
like the ADS1605 or ADS1606 from TI. You would still need the digital
circuitry to be able to adjust the front end gain, even with a 16 bit
converter, in order to achieve acceptable overall dynamic range. Dither
will help with the weak signals; just make sure your front end is
sufficently noisy (say, 1/2 to 2/3 LSB pk-pk noise). That is one of the
things I find amusing with analog to digital conversion - the presence of
a little noise can actually improve performance :-)

--
Greg

--The software said it requires Win2000 or better, so I installed Linux.

Hi Greg,

I seem to remember reading in one of the trade magazines that current digital designs actually make use of aliasing by
undersampling to reduce the required rate down to a more reasonable level. This makes sense if the base bandwidth is much
smaller that the carrier (e.g. 10khz vs. 1.6Mhz). You also don't need to preserve the fidelity of the carrier, you just need to
be able to detect the modulated signal.

Mark Robinson
"Greg Pierce" wrote in message news
Conceptually, it is not a bad idea - a broadband RF amplifier for a front
end, rolled off below 400KHz and above 1.7MHz, whose signal is then
digitized. The digitized RF is sent through the DSP, which does your
bandpass filtering and detection in the digital domain, and outputs the
"detected" signal into a conventional 16/44.1KHz (or less if you want)
DAC.

The main problem isn't the RF front end or the DSP - the front end can be
made either with tubes, JFETs, bipolars, or easiest of all with some good
high-speed op-amps (especially since we are talking 2 MHz here), and
sufficently powerful, inexpensice DSPs are readily available from TI, AD,
Mot, etc. The problem is the ADC. You will need at least 3.5MSPS - not
too bad - with sufficent dynamic range, which IS a problem. 10 and 12 bit
converters are readily available that can digitize at that speed, but 12
bits isnt enough; the dynamic range is only a little over 70db. 16 bits
with a good SFDR is what is really needed, since the incoming signal
can have a very large variation is signal strength - you dont want the
strongest to exceed the ADCs input range, while you want the weaker
signals to toggle more than a couple bits. 18 bits would be even better,
but good luck finding one. In the 16 bit realm, you could use something
like the ADS1605 or ADS1606 from TI. You would still need the digital
circuitry to be able to adjust the front end gain, even with a 16 bit
converter, in order to achieve acceptable overall dynamic range. Dither
will help with the weak signals; just make sure your front end is
sufficently noisy (say, 1/2 to 2/3 LSB pk-pk noise). That is one of the
things I find amusing with analog to digital conversion - the presence of
a little noise can actually improve performance :-)

--
Greg

--The software said it requires Win2000 or better, so I installed Linux.

Fred Nachbaur wrote:


[...]

Thanks, Michael (and everyone else). Seems I have indeed been "out of
the loop" too long, as regards the capabilities of present-day digital
technology. (Heck, I can remember when an 8 MHz clock was screaming fast!)

I also received a fascinating reply by email, I'll try to convince him
to post it here because it truly was an eye-opener for me.

This post may be in poor form both because I'm "replying" to myself, and
because it's somewhat off-topic. However, I found the content below very
fascinating, and hope that at least some others will also.

This has been copied from a private email conversation with Geoffrey
Rochat, with the author's permission.

quote
In fact, what this fellow's talking about is already all the rage -
although with semiconductor front ends, not tubes. One way to do this
is to undersample the incoming RF with a fast A/D, essentially creating
sum and difference frequencies "inside the A/D", then performing
filtering and demodulation magick {sic} inside an hellaciously-fast DSP.
Now available off-the-shelf from at least a half-dozen companies for a
couple of bucks.
Another way out is to mix the incoming RF with the output of a
quadrature local oscillator at the carrier frequency, giving I and Q
baseband difference signals (and sum frequencies, of course, but we can
punt those), which are then sampled by an A/D at greater-than-Nyquist
frequencies (double audio frequencies), then performing other filtering
and demodulation magick inside a hellaciously-fast DSP. As to which
method to use..., well, if you think the battles between whether a
concertina or a cathode-coupled phase splitters is best are rough, you
should hear these! In any event, do a Google search for
"software-defined radio", then stand back. Also too, the ARRL (and
probably the RSGB, too) have a lot of information on the topic. And, no,
I'm completely lost by the math.
/quote

My reply:
"Thanks for the info. Wow, guess I must have been "out of the
loop" as it were, for too long."

quote
Fred, I'm a consulting electrical design engineer by trade (Business is
*rather* slow these days...), and I've just finished designing,
implementing, debugging and now documenting, which activity this e-mail
is helping me to procrastinate upon, a special-purpose FPGA containing
something like 250K gates, that performs roughly two score simultaneous
processing functions on 32 bits of data arriving at 100 million words of
data per second, one word per central clock tick. And my client is
complaining, after the fact, of course, that it's too slow for what he
wants. To which I say that I'll be de*LIGHT*ed to double, even
quadruple, its capability IF HE'LL BLOODY PAY ME FOR IT!!!

I remember when you talked about a 100MHz "clock tick" you were
discussing the FM broadcast band - serious VHF stuff. And you needed
fancy things like lighthouse tubes to get there. These days 100MHz is
practically DC.

When I first started in this business, the laws of physics said that you
couldn't run more than 4MHz down wire-wrapped twisted pair. That's what
we did to clock Z80s, and that's was the best that could be done. Then
came 10Base-T Ethernet, and 10MHz down twisted-pair became OK. Then came
(and eventually went) 16MHz Token Ring. Then came 100Base-T Ethernet,
and 100MHz - 100Mhz!!! - down twisted-pair became a commonplace. And
right now the IEEE is finishing up its specs for 1000Base-T, wherein
they'll push 1GHz down twisted-pair. Indeed, if what Mr. Bush keeps
insisting is The Recovery hadn't taken hold, it too would be a
commonplace by now. My point here, and yes I do have one, is that what
really happened is not that engineers got cleverer, it's that the laws
of physics changed. I think it has something to do with Heisenburg's
Uncertainty Principle, or Schrodinger's Cat, or something, but by
observing high-speed signals in twisted-pair that somehow changes the
laws of physics to make that sort of signal propagation possible.

Well, it's a theory anyway, and I'll thank you very much not to go
poking holes in it with nasty facts.

So I think a similar situation entails with DSPs today. TI, Analog
Devices and a buncha other outfits I've never heard of (some of whose
names can't be spelled with the Latin alphabet, and can only be rendered
in binary) have el-cheapo DSPs from which they claim 1 gigaflop+
performance. I'd laugh, except that they're so popular no distributors
can keep 'em on the shelf. Barnum said you could fool some of 'em all
the time, and all of 'em some of the time, but not all of 'em all the
time, so I guess the DSP makers have to be doing something right.

BTW, another chip I did for that same client contains, as part of its
function, circuitry that measures the time interval between the edges of
two signals, with a maximum range of 24 seconds, to the nearest 20
picoseconds. Ayup, 20 x 10^-12 seconds. And the client is upset, 'cause
he was really, really, really hoping for 10 picosecond resolution, and
he doesn't think 20 picoseconds is good enough.

And my question is: How can he tell?

Anyway, this is why, in my copious free time, I mess around with tubes.
The mindset is entirely different. You can't bull your way through a
tube design, hogging resources, you've got to keep it simple. And you
can't shoot for "perfection", you've got to carefully work your way to
"good enough", properly defined, and not overdo it. That's why I have
to laugh at the golden eared crowd. Cripes, if they want "perfect"
amplifiers, buy something with semiconductors in it! Perfection is much
easier to achieve if you through a zillion transistors at a problem.
Throw a zillion tubes at a problem and all you're doing is making heat.
/quote

--
+--------------------------------------------+
| Music: http://www3.telus.net/dogstarmusic/ |
| Projects, Vacuum Tubes & other stuff: |
| http://www.dogstar.dantimax.dk |
+--------------------------------------------+

On Mon, 06 Oct 2003 14:21:23 +0000, the highly esteemed Mark Robinson
enlightened us with these pearls of wisdom:

Hi Greg,

I seem to remember reading in one of the trade magazines that current digital designs actually make use of aliasing by
undersampling to reduce the required rate down to a more reasonable level. This makes sense if the base bandwidth is much
smaller that the carrier (e.g. 10khz vs. 1.6Mhz). You also don't need to preserve the fidelity of the carrier, you just need to
be able to detect the modulated signal.

That seems to be an unlikely method. If you sampled at, say, 900KHz, then
everything above the Nyquist rate will get aliased. This doesn't seem to
bad at first glance - AM center frequencies are at 10KHZ increments,
starting at 460KHz (455KHz is the edge of the band). Thus, 460 will get
aliased to 440, 470 to 430, etc. The problem is when you get to 920KHz;
it gets aliased back to 460KHZ! 930KHZ is fine - it gets aliased to 465,
where the very high order digital filter can still seperate them despite
the frequency spread overlap. However, every 20KHz will get aliased
back on top of an existing lower frequency.

I suppose that if you are tuning below 900KHZ, you could precede the
ADC with a suitably high order Chebyshev low pass filter to get rid of
the frequencies above 900KHZ. Likewise, tuning above 900KHz would
require the use of a high-pass filter. This wouldnt really be any
different from the high-order low-pass filter you would need to use
as an anti-aliasing filter when sampling at a higher rate (such as 5MSPS).
You would use a high-order Chebyshev design (in this application, the
passband ripple in the Chebyshev is of no consequence) to rapidly roll
off everything above about 1.7MHZ. The sampled area between 1.7MHZ
and the 2.5MHZ Nyquist frequency would be a wasteland of aliased
frequencies caused by the finite rolloff of the Chebyshev filter, and you
would simply discard it (you don't tune above 1650 in the AM broadcast
band), but you would still need to sample at that high of a rate. Actually,
higher yet would be even better, since it reduces the burden on the analog
anti-aliasing filters. The TI converter I mentioned can actually sample at
10MSPS with 16 bits of resolution, which would allow the use of a lower
order filter.

IMO, since such converters are readily available at reasonable cost, and
implementing digital front-end gain control in the DSP is fairly trivial,
I would opt for that method rather than the undersampling method. I
believe the overall system would be less complicated...

--
Greg

--The software said it requires Win2000 or better, so I installed Linux.

On Mon, 06 Oct 2003 18:49:08 -0700, the highly esteemed Greg Pierce
enlightened us with these pearls of wisdom:

On Mon, 06 Oct 2003 14:21:23 +0000, the highly esteemed Mark Robinson
enlightened us with these pearls of wisdom:

Hi Greg,

I seem to remember reading in one of the trade magazines that current digital designs actually make use of aliasing by
undersampling to reduce the required rate down to a more reasonable level. This makes sense if the base bandwidth is much
smaller that the carrier (e.g. 10khz vs. 1.6Mhz). You also don't need to preserve the fidelity of the carrier, you just need to
be able to detect the modulated signal.

That seems to be an unlikely method. If you sampled at, say, 900KHz, then
everything above the Nyquist rate will get aliased. This doesn't seem to
bad at first glance - AM center frequencies are at 10KHZ increments,
starting at 460KHz (455KHz is the edge of the band). Thus, 460 will get
aliased to 440, 470 to 430, etc. The problem is when you get to 920KHz;
it gets aliased back to 460KHZ! 930KHZ is fine - it gets aliased to 465,
where the very high order digital filter can still seperate them despite
the frequency spread overlap. However, every 20KHz will get aliased
back on top of an existing lower frequency.

OK, I was asleep at the keyboard here - 920 would get aliased to 430, and
930 would get aliased to 420, etc. All of them would land on top of lower
frequencies.

Whoa, I haven't been getting enough sleep...

--
Greg

--The software said it requires Win2000 or better, so I installed Linux.

Heathkit used to make a totally passive AM tuner. Basically a tuned front
end and a detector. I used to use one of these with a preamp on the output
for a signal injector on my workbench. Audio Quality was quite good as AM
broadcast went anyway.


"Michael A. Terrell" wrote in message
...
Fred Nachbaur wrote:

Eric S. wrote:
I've been eagerly following the long-ongoing discussion on building a
high-quality AM tube tuner which would excel at both sensitivity and
audio fidelity.

It got me to thinking that maybe a tube/digital non-superhet hybrid
makes sense, to combine the old with the new. However, before I
present my idea, let me state very clearly that my knowledge of radio
circuitry is very minimal, so the idea may be beyond silly, or it's
already being done in some incarnation. So, be gentle with me, LOL.

The idea is after the first-stage tuning which uses a tube-based
circuit for low-level amplification, to digitize the weak signal
rather than mixing it with IF as is done in a superhet, then use DSP
techniques to extract the audio from the signal (don't know if one
would digitize the original signal itself, or to somehow subtract the
carrier before digitizing.)

I have no idea if it is possible to accurately digitize such low
voltage signals (at least in a practical sense). But if doable, I
surmise that clever design may meet the various requirements of
good sensitivty, selectivity, dynamic range and audio fidelity.

Again, it is patently obvious that I border on clueless on this
topic, so feel more than free to set me straight.

Eric S.

Hi Eric,

Well, I think you have a clue, but not much beyond that. ;-)

In the spirit of continuing mental exploration of your ideas, here is
what you'd have to do to build a fully digital AM radio:

There are essentially two operations that have to occur: (Actually
there's a third one; amplification. But I'll leave that one alone,
assuming that we'll be using a "suitable" broad-band preamplifier of
some sort.)

1) Tuning
2) Detection

The first operation separates out the one station you want to listen to,
The whole idea of the superheterodyne is merely to make this operation
easier, in that the main amplification only has to be done at a single
frequency rather than having to track multiple amplifier stages as in
the TRF (tuned RF) design. The second operation demodulates it so you
can hear it.

It's conceivable to do the tuning operation using DSP. But the sample
rate would have to be very high - at least twice the maximum frequency,
which translates to over 3 MHz. (As a point of comparison, hi-fi audio
is now done at 96 kHz, i.e. you'd have to sample over 30 times as fast!)

Then you'd need a software bandpass filter using FFT to select the
frequency range corresponding to your station of interest.

OK, now you've got the modulated RF. You have to demodulate it. This is
not a matter of subtracting out the RF, since the RF and AF are *mixed*,
not *added*. Rather, you'd have to simulate the detector in a
conventional set, by introducing a non-linearity. The simplest would be
to "chop off" the top or bottom of the waveform, and apply a low-pass
filter (again using FFT software).

Finally you'd have to restore the DC level, which can again be
accomplished in software.

But -- can this all be done in real time? I have my doubts. Perhaps with
a big Linux cluster or other distributed computing system you could
approach real-time processing at this speed, but even that would be a
chore. (I can see it now... 100 X-boxes linked together to digitally
simulate an AA5.)

Rather like using a crane to pick up a pearl. Possible, but fiddly.

Cheers,
Fred
--

Fred, the Microdyne RCB-2000 telemetry receivers I worked on had a 90
MHz IF that fed a D/A card. then it used a pair of FIR filters to select
the desired signal. Another pair of FIR filters followed that to shape
the output to the demod to recover the data. The IF bandwidth was
software selectable from 10 KHz to 40 MHz, and the "Video" filters that
followed worked the same. With the FIR filters, a lot of the processing
load was removed from the microprocessors, which were 16 bit Motorola
chips. It took more processing power to handle the GUI and to drive the
front panel display. It used a Cyrix processor, and ran under embedded
NT. There were six processor chips in the radio on a custom buss,
controlled by a 19.2 Kb serial port on the embedded controller. All
this, for just $80,000.00 US

--


Michael A. Terrell
Central Florida