Two ADCs in a box?

k3Tim

A posting here in the community lead me to wonder if there are two ADCs in a 6500.  To create the I/Q channels I would expect so. A digital trick Hilbert transform to make generate a Q from I doesn't seem feasible.  

In checking the docs about the 6500, it implies one ADC converter.

The radio obviously works as built, this is more a tech/edu question.

-.-  _..--
K3   Tim

Jim Gilliam

Isn't the Q signal in quadrature with the I signal? Seems this could be accomplished easily by taking the I signal and digitally delaying it 90 degrees.

Jim, K6QE

k3Tim

It is quadrature but one wants the actual phase information.  On the I/Q plane the I and Q components can be anywhere independent of each other.  In using the phase shift, they are orthogonal but one is predictable (One knows Q given I).

An analogy is distance and bearing.  If a ship gives a vector that has only magnitude one can't get the bearing from the bearing information only.  You know the vessels is 10,000 meters distance but where on that circle ?  (bearing)


Another analogy is averaging multiple samples to reduce noise.  You can't add the same sample 16 times and average it to remove noise,  The noise is random but zero mean, therefore added enough times, the noise reduces due it averaging out.

Thanks,

Joe, KQ1Q

An ADC of sufficient resolution sampling at the Nyquist rate of 2x the maximum desired bandwidth will capture all possible information below 1x that frequency. The Flex 6500 16-bit ADC samples at 491.5 megasamples/sec, so according to Nyquist-Shannon criteria it captures (and could perfectly reconstruct) all possible waveform information from 0 to 245.75 Mhz. That includes amplitude, phase and frequency.

I don't see how an extra ADC would add anything to this. If a duplicate copy is needed it could be cloned from the original sample set. It would be an absolute duplicate with all the original characteristics.

The 2nd ADC on the 6700 is useful for diversity reception where two separate antennas with different signals are sampled simultaneously.

Jim Gilliam

All I could gather from his comment was that perhaps the second ADC would capture the data 90 degrees shifted from the other. However, I can't see what would be the difference than just delaying a clone of the original ADC.

Joe, KQ1Q

The Flex spec says 300Khz to 77Mhz, so I don't know the basis for that unless it's oversampling for some reason, or maybe using low-pass filtering below the Nyquist bandwidth to avoid aliasing. But within that range, absolutely accurate information capture is performed. An analog signal reconstructed from that digitized data would be indistinguishable from the original analog signal.

I don't remember if there's a formal theorem for this, but I thought there was no limit to the transforms possible in the digital domain. IOW you never hit a wall and have to drop back to analog because digital manipulation wasn't possible. The 6500 has four slice receivers which are derived from a single ADC data stream. On the 6700 the additional four slice receivers (total of eight) are only because of computational issues, not the 2nd ADC per se. A sufficiently fast ADC and processing pipeline could produce eight slice receivers from a single ADC data stream.

Jim Gilliam

His argument was that he felt two ADC's would be needed to get a "correct" Q signal as opposed to the "correct" I signal. However, I got very confused when he said that there was a loss of phase information. I had to go back a review my sampling theory and you are correct about all the information is captured on a band limited signal that is sampled at twice the rate of the bandwidth. It never occurred to me that the phase would be lost, because if it was, then it would be impossible to reconstruct the original signal.

Jim, K6QE

Gerald-K5SDR

The FLEX-6500 has one ADC the FLEX-6700 has two independent ADCs. All I/Q mixers are in the FPGA.

Jim Gilliam

How is the Q signal "constructed" before the mixing?

Gerald-K5SDR

The ADC over samples in real mode. FPGA down samples and the implements complex I/Q mixers in the FPGA. it's all digital. No need for two ADCs unless you want to do diversity or receive on two antennas at the same time.

Jim Gilliam

For those mathematically inclined here is a good write-up on I/Q mixers:

http://rfic.eecs.berkeley.edu/~niknejad/ee142_fa05lects/pdf/lect16.pdf

k3Tim

Thanks for the comments and I hope I am not adding to the confusion or noise level.

I thought a second A/D would be needed to retrieve both I/Q channels in the direct conversion receiver.  Check out Mr. Campbell's architecture from 
https://www.arrl.org/files/file/Technology/tis/info/pdf/9804x040.pdf

Block diagram #1 shows the work being done at RF, generating the I/Q oscillator at frequency and two A/D channels.  

After further research and referring to "Digital Signal Processing in Communication System" by M.E. Frerking (aka. W0EQO) figure 7.31 there is a block diagram of a direct sampling receiver with a single ADC.  The I/Q is done in the digital domain where one can easily generate a cos(omega T) and sin(omega T).

In the same text, now chapter 4, the explanation of why the single ADC is sufficient is clear.  Using the author's example (except I will embellish it with Roulette Wheel) one can think of the signal as a rotating vector.  This is a different but equivalent way of working with the I/Q data.  If we can determine the rotational speed (RPM) of the roulette wheel and if it's going CW of CCW, we will know everything about it and hence the signal.  If every time the 7 hits the flapper an indication (IRQ or similar) is given we can determine the RPM from timing successive pulses.  To determine the spin direction, a second flapper at 90 degrees which also pulses is observed when the 7 passes this second flapper, then we can time the difference in the two pulses and know the direction of spin, CW or CCW.  If the second flapper was placed at 180 degrees, it doesn't allow one to determine the spin direction.

In previous post I said the Q data would be predictable given the I data and the fact the Q is 90 degrees different.  That's almost true for a CW signal.  The Roulette wheel example would be a constant RPM and direction of spin.  In that case the I/Q data would be constant / consistent.
With RF signals off the band and converting a large swat of frequency this is like observing 1000's of wheels and getting changing I/Q data with no correlation between the I/Q components.

I managed to wrap my brain around this simple though fundamental concept and hope this was helpful to others in the group. 

Regards,

Tim

k3Tim

Thanks for jumping in on this one Gerald.  Thankfully at US$150 per ADC we need only one...  

Best,

Tim

Tim - W4TME

That article describes a 1st generation, SDR that uses a quadrature sampling detector technology to do the phasing before the ADC.  In a DDC radio like the FLEX-6000, the antenna is essentially connected directly to the ADC and the phasing is done in the FPGA.

k3Tim

Thanks for the response....

The folks from Flex put on a good talk at the Yuma hamfest about this.  Anyone that gets a chance to hear the pitch should take the opportunity.

_..--
 TiM

N7CXI

I have no idea how Flex does it, but the CORDIC algorithm is often used elsewhere.

73,
Jim N7CXI

k3Tim

Hi Jim Thanks for the the link. Haven't heard of this and a Google search turned up a detailed PDF about implementation. Best Tim

N7CXI

Now that I'm more-or-less conscious, here's a link to some CORDIC references: http://www.dspguru.com/dsp/faqs/cordic FWIW, there are SDR implementations that use purpose-designed ADC's that have both I and Q outputs. My knowledge of those systems is dated, so I don't know if that kind of design is still viable or not, or available in the general market. HTH, Jim N7CXI