Practical Measurements on the 2nd IF

Warning

If you decide to use any of the information here, you will make your PICASTAR non-standard. Even if you wish to do the changes on a 'new-build' STAR, please build it first using all the standard build information, with no changes from the published design. Also you need to calibrate it using all the excellent BASIC programmes, to verify that it works correctly, before you modify it. This is what I did myself.
If you do not 'RTFM' (read and use all the PICASTAR documentation in the Yahoo! Groups Pic-a-Project group before you modify your build in any way, then no-one will be able to help you to get your own build working. You must join this group in order to obtain (and be permitted to use) the design information that you need. If you then need help (with an un-modified STAR) you will get it by posting your problem on the picastar-users group. If your STAR is modified in any way, then the very helpful people on the picastar-users group will probably not be able to help you.

Introduction

Glenn VK3PE and Roderick VK3YC both showed interest in my ramblings. Very fortunately, both Roderick and Glenn have built N2PK VNAs, and even more fortunately Glenn found that, in spite of the small coupling transformers, the VNA will work well down to below 15kHz. Below about 10kHz the performance suffers, causing harmonics rather than fundamental to be produced, wrecking the results. But it is good enough to see what happens in 'real' circuits.

It's always encouraging to see 'real' results from proper components, no matter how convincing the simulation results seem!

Both Glenn & Roderick have lashed up the 'original' and 'modified' diplexer/filter circuits (not the complete 2nd IF amplifier) for measurement in isolation, and have kindly allowed me to present their findings.

Glenn is also now constructing his 'n'th Star (a ComboSTAR of course) and has been able to measure the characteristics of the entire 2nd IF circuit from diplexer input to op-amp buffered output, using the original published circuit. Since then, he has rebuilt it with the new values, has again characterised it and I have included the results here.

The Diplexer / Filter

You have already seen the simulated performance of these, so here are some real plots.
Just as a reminder, this is the part being tested. The 220nF C631 is not included:
Schematic of the diplexer being tested

Before and After 'the mod'

Here is what Roderick's measurement setup for the original diplexer alone looks like:
Rodericks test setup

And the plot of this is as below, together with the modified diplexer.

Both the original (blue) and modified (pink) results for transmission are shown, together with the load impedance and resistance for both configurations. For the modified circuit lashup, Roderick used 200nF in series with the 47R 'HF Arm' load; the result is not much different from the suggested 100nF - impedance at 30kHz is better, but gain flatness at 15kHz is worse.
As far as I am aware, Roderick has not used close tolerance capacitors in either lash-up.Roderick's superimposed original & modified diplexer
Recall that the purpose of the circuit as a diplexer is to terminate the mixer with the correct impedance (50R) at all frequencies, especially those to be expected from the mixing action.
At 15kHz, the 'difference frequency' and the wanted 2nd IF, it is not bad, just before it heads off for some rather high value. At the 33kHz image it is en-route for something big.

Easily seen above is the same pattern as predicted by the simulations, with the load impedance of the original 'diplexer' (purple trace) dropping off the scale before 2MHz. This negates the presence of the HF termination arm and shorts the 2nd mixer output at higher frequencies, exactly as predicted.
So much for maintaining proper termination at the important frequencies of 1st IF (10.7MHz), 2nd LO (10.7MHz) and the sum frequency (21.4MHz). Rather than 50R (actually the 47R R601) it is practically zero. Ooops!

The lack of attenuation at the first aliased frequency (33kHz) is also clearly visible on the transmission trace (upper blue).
Recall that the only purpose of this circuit as a filter is to block any aliasing frequencies from reaching the ADC. If they get to the ADC, they will be down-shifted by sampling into the normal IF region and subsequently indistinguishable from genuine IF signals. In STAR the first alias occurs at 33kHz since the ADC sampling frequency is 48kHz (alias is at 48kHz - 15kHz). Oh dear. Not any significant attenuation, maybe a dB or two?

It is painfully obvious that the original diplexer/filter in the PICASTAR 2nd IF design does not do either of the things it should.  It is neither a diplexer nor a filter.

In contrast, the revised diplexer/filter shows good cutoff at 33kHz (pink trace) and the expected 47R termination impedance at high frequencies (up to 11MHz shown).  The higher insertion loss is fortuitously more than compensated for in later stages by using proper-value coupling capacitors in the modification.

The marker 'Mkr 5' at 30kHz is at the lowest possible frequency component for aliased signal. Anything below this is rejected by the DSP algorithm. See here for more on this topic. The LO is shifted between USB and LSB, to keep SSB straddling the 15kHz IF, so that the wanted SSB is 15kHz ±1.5kHz and the first image, which will alias down to the 15kHz DSP passband, would be at 33kHz ±1.5kHz.
Even with these somewhat pessimistic markers, it is easy to see from the curve that the original filter attenuates the first image by less than 2.5dB; the modified filter by at least 7.6dB. Later, once the LF rolloff of the various coupling capacitors in the remainder of the circuit takes its toll, the original 2nd IF actually has more gain at the first image than at the desired 15kHz (see below)!  Remember that the above composite plot is for the Diplexer/filter alone.

Even on it's own, without the rest of the second IF circuit, the original diplexer/filter does not behave at all properly in any respect.

Input impedance

Roderick has produced a very useful Smith chart showing the input impedances of the original diplexer and the recommended modified 'T' diplexer, with both 100nF and 200nF capacitors in the HF arm.
This is from actual measurements of 'lash-ups' of the diplexer/filter sections alone.
The tighter control of input impedance with 200nF in the HF arm is easily visible, but this occurs by the HF arm 'coming in' at a too low frequency. 200nF presents shunt loading in parallel with the LF arm at 15kHz, worsening passthrough slope and attenuation, so is not recommended.

Smith chart comaring diplexer input impedances
This Smith chart clearly shows how, in the original diplexer, capacitors C639 (10nF) and C651 (100nF) certainly short out the mixer at 10MHz. But at 10MHz the two modified diplexers both end up at 47R, the resistive value of the HF arm terminator R601, as was obviously intended.

What about the whole IF?

Not quite so easy to measure the 'before' and 'after' of the complete circuit from "LF-PORT" through to C662 - at least that was what I thought until Glenn told me that he had already built this section using the original components (as per his ComboSTAR circuit) and intended re-building it with my recommended modifications.
So Glenn set about isolating the 2nd IF circuit and connecting it to his VNA.

Because the 2nd IF amplifier has gain, he inserted a 30dB attenuator at the input and a buffering 6k8 in series with the output.  This 6k8 is needed also because the op-amp cannot drive a low resistance load; it is specified only down to 2k load even with +-15V supplies; we have a single 10V supply.

Here is a diagram of Glenn's test configuration using his VNA and latest ComboBoard (thanks Glenn!)
Click image for a pdf:
Glenn's Combo IF2 test configuration

Glenn tested his original 'standard' second IF  Receive path, then modified it and tested again. The 'unmodified' test setup was not exactly as shown above; the 30dB attenuator was at the 'Bridge Out' when performing transmission measurements, so was not included within the calibration. You can see this in the plot below.

Glenn's original complete 2nd IF, incorporating all the Official Modifications but none of my recommended changes.
The turquoise trace is spurious, and the dark blue trace is the remnants of the plot with no 30dB input attenuator as was used to obtain return loss. Glenn subsequently calibrated with the pad in-circuit as shown in the diagram above. I will replace the two plots below soon. The pink trace shows clearly the LF rolloff due to the much-too-low coupling capacitors. See below for more information.
Glenn's whole Original 2nd IF characteristic

In the above plot, which includes every component in the 2nd IF amplifier from the diplexer input right through to the op-amp buffer output, the devastating effect of the too-low coupling capacitors is easily seen as LF rolloff.
This was mentioned previously as worsening the rejection of the 33kHz first image from about 2.5dB in Roderick's mock-up of the diplexer alone - here it is easily seen that the rejection at 33kHz relative to 15kHz is not only less than 2.5dB, but it is in fact a relative gain of 0.3dB! The response to the first image is stronger than that of the wanted second IF. This is slightly worse than the simulation result, possibly due to component tolerances (the large ceramic capacitors are rather widely toleranced, both in the filter and the couplings).

Just to see how closely Glenn's measurements correspond with the simulation, I have scaled and offset the simulated transmission curve to correspond with the plot above, and superimposed it. The dotted axes are from the simulation, lined up on frequency and arranged to align appropriately in amplitude, bearing in mind the differing scale divisions. The dark blue line is the simulation. Some fragments of other simulation output can be seen and should be ignored.
Superimposed actual & simulated whole 2nd IF plots 
 
The similarity of the simulation (dark blue) and real transmission (pink) curves is obvious and gratifying! Remember that the gains shown are modified by the need to avoid overload of the amplifiers in the circuit.
A slight frequency offset and some shape difference is apparent; no doubt due to real component tolerances.

In particular, the inductor (RFC612, 330uH) and associated capacitor (C639, 10nF) and filter output capacitor (C651, 100nF) may not be close tolerance, especially the 100nF which, in the parts list, is just the same as the decoupling capacitors used extensively on STAR.

I don't know what Glenn used, but for my own build the 330uH was ±5%, and the 10nF was ±5% NP0, however the 100nF was 'just an ordinary' 100nF 0805 from Farnell, reference 141-4664, which is "Dielectric Characteristic:X7R, Capacitance Tolerance:± 10%, Temperature Coefficient:± 15%" - nowhere near adequate for a tuned circuit! Even then, I had chosen a well-specified version of this type of capacitor; for many constructors the 100nF will have been part of a capacitor kit from China, OK for decoupling but do you know what characteristics it actually has?

See the full simulation results for better understanding.

Glenn has now modified his 2nd IF to my redesign and measured the IF performance, so we actually have the 'before' and 'after' measurement results. Glenn's modification is physically a little different from mine, having made good use of a fortuitous via on his PCB to avoid standing the 330uH inductor upwards on surface-mount pads, as shown here:

Glenn's ComboBoard showing the modified second IF circuit
Glenn's modified second IF.
The big electrolytic in the left foreground is temporary, for measurement only, to block the d.c. bias at the bus switch from flowing through the VNA source.

Glenn notes: "My inductors are as close as I could get.  As per usual, I bought a bunch of chokes and they are all over the place. Best I can find is 662uH and 351uH. The others are way off. That's the only potential downside here - getting inductors that are close. Mine are local store parts, not Farnell. The 100nF caps used are all NPO's and all very close to 100nF. Very close."
Just for the record, Glenn's selected inductors are -2.6% and +6.4% of nominal.

Glenn has measured and selected his inductors from local supplies. I used Farnell 5% parts which have proved to be excellent (albeit near the +5% end). Also Glenn has used the rather expensive close-tolerance NP0 100nF ceramics as I recommend. Cheap, lossy, bad tempco X7R or worse dielectric decoupling capacitors are simply not worth using in tuned circuits - even if you measure & select values.
May I re-iterate that it is no use having a filter that is expected to have a precise characteristic if it is fed with poor components?
Do not spoil the ship for a ha'porth of tar (do non Brits understand this one?).

Glenns 2nd IF performance, before & after the modifications

Roderick has produced a Smith chart of Zin from Glenn's VNA measurements. As he points out, please note the 'curl' at the low frequency end of the plots, which is due to the N2PK VNA running out of performance.

Smith chart: before & afer mod

The 'before' and 'after' overall plots of Glenn's Combo-Board. Remember that he used a buffer resistor between the op-amp (IC606B) output and the VNA to avoid under-terminating the op-amp. This gives the apparently low Gain figures that you see:
Before & After measurements on Glenn's ComboBoard IF2.
As you can see (above), the gain of both circuits behaves in the same way as the simulations. It's good to see them together on the same plot; it strikes home the relative performance and avoids one concentrating excessively on the incremental gain slope of the new filter around 15kHz!

Which of them properly (or even actually) attenuates the 33kHz quantisation image?
Which one properly terminates the 2nd mixer at 15kHz, 10.7MHz and 21.4MHz (1MHz is the highest actually shown)?
Which one is what you would expect in your PICASTAR?

You can clearly see (dark blue plot) that the attenuation of the original filter at 33kHz (first image frequency) is not much different to that at 15kHz (the 2nd IF), showing even a slight relative gain at the image, which is not at all useful.
By contrast, the 'after' light-blue plot shows a steep cut above 20kHz; from the markers Mkr3 and Mkr5, which represent the gains at 18.8kHz & 29.5kHz, pesssimistically less than the span of actual IF and image frequencies,  the cut-off achived for the image is 8.7dB, which will suppress the image to the point where it will not be troublesome.
Remember that this image (at nominally 33kHz) is almost entirely formed from noise in the circuitry following the crystal filter. 'Real' signals will have been sliced away by the roofer. But at maximum gain (i.e. weak signals) this noise would otherwise be aliased by the ADC sampling, and will then add statistically to genuine noise at 15kHz and be processed as genuine input. Whether this worsens the noise floor measurably depends on the relative contributions of this noise and front-end noise, but having originally included a filter in the design, it seems that it would be rather nice if it worked as required, instead of - not.

Again, thanks to Glenn for producing these plots and allowing me to include them, and to Roderick for presenting and annotating the graphs.

From both the above plots, the improvement in termination of the 2nd mixer can easily be seen.
The pink 'before' trace, 'Orig Zs' is alreading diving towards a dead short at 1MHz; it should be 50R at 10MHz and 21MHz in order to have any purpose. The HF arm of the diplexer is bypassed by the LF arm capacitance, as predicted somewhere here.
The black 'after' trace, 'Mod Zs' shows the enormously improved load impedance, both at the 2nd IF (15kHz) and 1MHz, where the 47R termination of the HF arm of the diplexer now works as originally intended. This HF arm continues to be correct (as correct as a 47R makes it) right up to way above the mixer sum frequency, normally 21.4MHz. This is also shown well on the Smith chart.

Another plot of a 'real' filter has kindly been produced by Steve, G7WAS. This is for the passive filter in the 'receive' direction and does not include any active elements.

Steve's filter plot
The cutoff above the desired (15kHz) IF is clearly far better for the Modified filter.

A wider frequency range plot shows the higher frequency behaviour:
Wider plot of filter from Steve G7WAS

These were done 'the hard way' by Steve, measuring many, many points in order to achieve the excellent plots. The 'features' of the original filter (passing the alias frequency and re-entry after the dip) are shown clearly by the plots and are very similar to the simulations, thankfully! Although the new filter is not completely flat at 15kHz, the gain error over our narrow bandwidth is small.  Steve intends measuring the response of the entire 2nd IF receive amplifier sometime.

Potential for problems

Roderick VK3YC has pointed out the oppsite slopes produced by the original & modified filters, shown below as plots of Glenn's 'before and after' measurements but also clear from the simulations.

It is apparent that the original filter has a rising response across the 15kHz 2nd IF, while the modified filter has a falling response (the bandwidth used for SSB is around 2.5kHz centred on 15kHz AFAICT). It is just conceivable that the DSP flattens the original response, rather than assuming it is level. If this is so, then the correction applied to the new filter will make the now falling response 'twice as bad'.
This certainly needs investigation, although I cannot believe it would be so, because:
For the above reasons, I cannot believe that the response has been considered as anything but level. If this is so, then there will be no compensation in DSP  (other than maybe sin(x)/x correction) and the reversal of slope will have no consequence. One day I will look and see and comment here.

In the meantime, you may like to ponder on my Caveats and suggestions. I can give no guarantees, you are welcome to my ideas (which are given in good faith), but your STAR may become 'non-standard' and stop working properly.

Your Comments

Now that you have seen what the 'before' and 'after' of Glenn's real PICASTAR 2nd IF does, you may wish to let me have your comments. If you do, and if you give me permission, I will put them here.

Glenn VK3PE: "Hi Bob, very happy so far with the mods and they will be staying there !!"
Roderick VK3YC: " If I had the modifications in my PicaStar I wouldn't be removing them, because I believe the modifications are correct."

Credits

I am very grateful to Glenn VK3PE and Roderick VK3YC for the interest they have shown and for their patience in producing the measurements I have included above (and loads more before that). Without their help I would have been unable to provide such good information. Also thanks to Glenn for allowing me to use portions of the schematic in order better to illustrate the affected circuitry on this and other pages.