[hpsdr] IQSD - was RE: hOw AbOuT a PiNG?

[Marco IK1ODO -2]

At 16.07 17/04/2007, Gerd, DJ8AY wrote:
>Ahti and Marco,
>
>thanks for the info.
>Up to now I do not know anything about TINA software.
>I will try to find information on it.

Texas Instruments web site, 
http://focus.ti.com/docs/toolsw/folders/print/tina-ti.html
It is an interesting spice-like simulator, and does transient and 
noise analysis.


>I mainly wondered if you are using buffers between the switches and the
>OPA1632.

No, the focal point is the matching between the source impedance of 
the mixer output (220 Ohm in my last circuit) and the noise impedance 
of the opamp, to have a good energy transfer to the amplifier. I 
worked a lot on this point, and wrote an excel program to evaluate an 
opamp's noise figure in relation with the source impedance. I wanted 
to have a quiet (and simple) frontend, and the OPA1632 is the best 
choice that I found (found by Ahti, in effect!). It is cheap, very 
linear, has enough band and suits ideally the differential input of 
the professional audio interfaces. Any kind of buffer before the 
OPA1632 simply makes things more complex.

>I also have some general questions on the QSD and hope to get your help:
>
>1. which criterias lead to the choice of the sampling capacitors (here 5n,
>with SDR1000 15n ...) or what happens if you use a much smaller and a much
>higher value?

There is no sampling capacitor ;-) since it is not a sampling 
detector "à la Tayloe". The name QSD is misleading! The 
switch/diplexer group is a source with his Z that injects current in 
the virtual ground of the opamp. The integration is done by the 100pF 
cap on the opamp; the diplexer simply avoids that out of band energy 
(over 175kHz in this case, that is the crossover point of the 
diplexer) reaches the opamp. The 5nF capacitor is not a sampling 
capacitor in the usual terms. I wanted to have a wide RF bandwidth, 
to be able to receive +/- 96 kHz with a 192kHz sampling rate, so 
choosed a cross point about twice that. The integration capacitor is 
made small, again to be flat at the upper limit of the response (and it is).


>2. which influence has the timing of the switching pulses (overlap, gap,
>delay ...)

Ah, this is a long discussion. My first mixer had 90° switch 
conduction angles, so no overlap. It worked very well up to 20MHz, 
with some degradation at 28 MHz. I could not test it at 50MHz (no 
fast dividers available). Then Ahti found that by simulating it in 
TINA the recovered voltage was greater using 180° conduction, like in 
the SoftRock mixer. I was in doubt, so I did my simulations, and saw 
a possible side advantage; using a wider conduction angle the real 
gate opening time is greater at high frequency (50MHz). Effectively 
the 180° mixer seems to work better at 50 MHz; at lower frequencies 
the difference is marginal. But it also has one less IC, so it is 
simpler. From an energy point of view, in the 90° version there is 
always a switch open; in the 180° version there are two switches 
open, and the Z is lower, but in any case all the available energy is 
transferred to the load (the integrator). With the 180° switch the 
integrator C has more time to charge, but the source V is lower, so 
at the end there is no much difference in output voltage. I may send 
you the TINA models to play with, I found it instructive.

I'm still working on it, I have two things to try:

- increasing the mixer input Z to 200 Ohm like my previous version, 
and the diplexer Z to 400 Ohm; I suspect that this could reduce the 
switching noise.
- building a test jig to measure the switching delay and transients 
of the FST3125. I think we still have to understand it... now I have 
access to a 8GHz sampling 'scope, and I want to use it.

I started working on this circuit in Sept. 2006. After seven months, 
I still have to understand it completely; it is only apparently 
simple, but difficult to optimize. Something may be done by 
simulation, something needs experimenting, and it takes a lot of time 
(I have a family and a QRL too, HI).

73 - Marco IK1ODO


 1176828610.0