PICASTAR Second IF (diplexer and amplifiers)

This is a draft of the final form of this and the associated pages. Please let me know of the gibberish and broken links that are sure to have sneaked in while I was looking the other way.

I will be adding to and amending these pages as time goes by.

Glenn VK3PE has kindly extracted the modification details and bundled them into a downloadable pdf document, (1.6Mb) that might be convenient. This was updated on 7th July 2010 to amend a wrong reference and add a missed bias-resistor. Please let me know of any further errors or omissions in this.

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.

Foreword

Recently retired, I chose PICASTAR as a first and only rig to build mainly because I wanted to construct a mature design, completely 'done and dusted'. I was particularly attracted to the use of DSP, it was clearly the result of a great deal of work and is absolutely the right way to go.

Although it would be possible to ignore the shortcomings of the Second IF, and 'carry on regardless', it is perfectly apparent to me that if the original design team had known about these problems, they would have cleared them all up long ago. Therefore I can only assume that they will be happy to see these gremlins put to rest and not see my efforts as any form of criticism of their own excellent work; it is not.
I can't think of any good reason to leave in place design errors that were simply not noticed before.

There will be 'new' builders of PICASTAR, probably for many years to come. After all, I am one of them!  If they can build their STAR confident that no-one is turning a blind eye to problems, no matter how late in the day they are seen, this surely cannot be a bad thing?

Although the STAR as a whole works well without this attention, it is likely (but not inevitable) that purging it of obvious and correctable defects and deficiencies will improve things.  I have a personal opportunity to address any such things as I feel necessary now, while I am constructing my ComboSTAR (Thanks Glenn!) before it is all shielded and cased.

I am pleased and honoured to be a builder of the excellent PICASTAR by Peter, G3XJP.  Peter has produced a very advanced set of DSP code to process signals from the 2nd IF, with excellent results. This IMHO deserves being fed with high quality input from the analogue circuitry; on my own STAR I am not averse to seeking solutions to what I might perceive as limitations of the existing analogue & RF areas to feed the DSP with the best reasonably available 'food'.

If you were told that the whole 2nd IF circuit has problems, and that the diplexer at the ouput of the 2nd mixer has serious shortcomings, would you be surprised?
Why have components installed that are doing nothing?
Did you expect the diplexer to terminate the 2nd mixer in 50 Ohms? At which frequencies? 
Did you think that the 2nd IF transmit buffering is linear?
    So did I until I did the sums.
Read on, to  see what I found, or take a sneak peek at some real measurements.

Thanks to information from Dave G3SUL on the anti-aliasing present within the CODEC, I no longer regard the anti-aliasing performance of the Diplexer/Filter as being a critical requirement. Nevertheless, there are other reasons why there are problems in this area, so I have used as a design target the provision of anti-aliasing performance.

The Purpose of the Diplexer and Filter

The sole purpose of the diplexer is to separate the wanted frequency and unwanted frequencies that emerge as mixer products from the 2nd mixer, and properly terminate them at the impedance of the mixer. The wanted frequency is then passed onward as the 2nd IF signal, and the unwanted frequencies are 'dumped' in the terminator.

It is intended to work by providing both LF (2nd IF of about 15kHz) and HF (all other frequencies) paths, each terminated at the 50R circuit impedance, so that the mixer is properly terminated at all frequencies. The LF path is through the inductive arm. The HF path is through the parallel arm consisting of 10nF in series with 47R, which simply shunts the input with a terminator at high frequencies

As a secondary (but important) function, a low-pass filter in the 'wanted' path is needed in order to remove in a secondary 50R path those frequencies above the normal 2nd IF (ours is 15kHz) that would otherwise be digitised by the ADC in the CODEC. In STAR there seems to be an attempt at incorporating this in the Diplexer. Because the DSP section uses an AC97 spec CODEC, this also has deep rejection of ADC aliasing frequencies, so it is likely that additional anti-aliasing filtering is unnecessary.

One problem with the original filter / diplexer is that it cannot be made to work as obviously was intended as it stands. The architecture of the filter causes it to present a capacitive input at high frequencies, so it shunts the input with a low impedance at the frequencies where it should be very high (10.7MHz and 21.4MHz). This bypasses the 'HF termination arm' of 10nF + 47R completely, and mis-terminates the mixer. This behaviour cannot be cured easily.  In addition to this, it fails even as an anti-aliasing filter, since the rejection at the first alias frequency is negligible (some of you will even have a higher gain at the first image than at the main frequency). This is compounded by the general a.c. couplings in the circuit, which begin to roll off the LF response above the 15kHz 2nd IF. This is probably just a mistake, but it exists in your STAR 2nd IF amplifier...
The DSP and aliasing
The Digital Processing will be happy to be fed with any digitised input that spans from d.c. to well over the 2nd IF; the software will do the rest (think of the amazing selectivity of a SoftRock or other SDR receiver, all due to DSP). The only possible requirement for our Filter is adequately to strip away the frequencies that the ADC will alias (fold back) over the top of the 15kHz passband of the DSP algorithm. These are all above the Nyquist frequency.
Please refer to the comment made by Dave G3SUL, pointing out that the AC97 CODEC used in the DSP section will also have built-in anti-aliasing filters.
Nevertheless, if we are to design a diplexer that will block the 1st IF, it makes some sense to consider it's performance as an anti-aliasing filter also.
Receiver Bandwidth
Recall that it is the DSP that provides the ultimate receiver bandwidth. This can be wide SSB, narrow SSB or CW as selected by you. Even AM or FM demodulators could be available; they are just DSP code. Other, prior bandwidth reductions are not there to give the final selectivity, but to remove unwanted signals that would compromise the real one (by intermodulation in amplifiers and by aliasing in the CODEC ADC). The 1st IF at 10.7MHz or thereabouts is to remove mixer image frequencies and large adjacent signals, using the roofing filter as the provider of selectivity. Neither 1st nor 2nd IF determines the final signal bandwidth of the Receiver; they just clean up the signal in ways impossible for the DSP to perform.

I am neither stating nor implying that these other elements in the receiver either work correctly or not. That is a separate consideration not dealt with here.
What to do?
I have not only analysed the existing design and found shortcomings, but given ways of removing these shortcomings and getting the characteristics that you thought were already there - with no copper mods or daughter boards, and all easily possible on Glenn's super ComboBoard professionally made PCB.

My findings (and solutions) are presented here in these pages for you to study; if you wish to comment please do so, and if you would like to adopt any of my 'unofficial modifications' please feel free to use them (of course, I give no guarantee). Anyone who has already screened and cased his STAR may have little use for these ideas. I am not encouraging anyone to modify a completed STAR, or even to adopt any of my suggestions.  But at least see what you might do.  Seriously consider at least performing the minimum changes on and around the 2nd IF Transmit Buffer to prevent peak clipping.

I have here proposed mainly some component value changes, with a few more complicated changes. There is no track cutting to do and no 'daughter boards' to provide. All the wrongs identified so far on the orginal circuit are corrected. It is all easily reversible.
The question is, has it done any good? Jump to here to find out.

History

PICASTAR 'inherited' the original 2nd IF design from the design of the DSP-10 by Bob W7PUA, to which the circuit is almost identical in appearance (the capacitor across our RFC612 is a different value). It seems as if Bob, in his turn, took some of the design from the R2 receiver by KK7B.  This has cloned not only the good points but also the quirks.  Although it functions, there are several areas of this circuit which are less than satisfactory and which I have addressed here.

Originally PICASTAR used a diode double-balanced mixer module to mix the IF (normally around 10.7MHz) and the crystal-controlled 2nd LO in order to produce the Second IF of 15kHz, for digitisation by the CODEC and subsequent processing by the DSP block.
Later this migrated to an add-on switching mixer using a bus switch FST3125 in a similar way to the First Mixer on the Magic Roundabout (MR).

A requirement of the diode double-balanced mixers is that all ports need to be terminated in the correct impedance at all frequencies that can be produced. The output port must be terminated properly  at the incoming IF (IF1), the incoming 2nd LO (LO2) and both major mixer products (the wanted one usually at LO2-IF1 and the corresponding unwanted frequency at LO2+IF1). Only by terminating at least these at the correct impedance will the performance of the mixer be maintained.
The more recent switching MixAdaptor is no less critical in this respect.

For constructors who, like me, have used the excellent PCBs from Glenn VK3PE, there is in any case no real choice but to use the switching mixer; the boards having been designed with only this circuitry present. This should produce better performance especially in IP3, and it is certainly cheaper in parts than a diode DBM module.
It is also very likely that those constructors who began with the diode mixer will have used the MixAdaptor to adapt their STAR to the switching technique, since it is one of the recommended modifications. Nearly every PICASTAR builder will use MixAdaptor in one form or another.

ComboSTAR

I am currently commissioning a PICASTAR ComboSTAR, and have developed an 'unhealthy interest' in trying to understand "why" in any areas that take my fancy!

You may have seen my analysis of the signal level in the First LO, together with a recommendation for component value changes that will ensure adequate ouput level.
For no reason other than that it seems to have a set of 'ad-hoc' modifications, I have looked more closely at the diplexer and amplifier that follow the second mixer. It caught my eye because, on the ComboSTAR schematic, some component values and topology seemed to have altered from the original design:


Diplexer fron Glenn's ComboBoard

I have been warned that there has been some "tuning" to the circuitry in this area, and that the GMS code in the DSP has been fine-tuned to work with the design as is, with the warning that it might not work properly if the circuit is not built 'as published', but I still cannot let so many design issues pass by.  If, on rectifying the problems, the DSP code seems to give trouble, I guessed that that might be my next port of call. Two wrongs don't make a right! In fact, the DSP and setup procedure work perfectly after the changes, which doesn't surprise me at all.

Why I looked closer in the first place.

There is a PI attenuator (pad) consisting of RZ2, RZ3 and RZ4 (in the schematic above) between the diplexer and common-base amplifier TR603 which has been 'doctored' to give a higher signal level by increasing the input resistor from 82R to 470R. The pad looked as if it originally had input & output resistances of 50R, for proper termination of the diplexer and matching to the amplifier emitter.
But changing the input 82R 9RZ40  up to 470R means that it no longer presents 50R to the diplexer LF port. Instead, the load seen by the diplexer at 2nd IF (15kHz) is actually nominally 102.4R!
This caught my attention as warranting a better understanding, which I will share. As I studied deeper, the real situation became apparent...  As always, I accept that you may disagree with my calculations, statements and conclusions - if you disagree, or even if you fully concur, why not contact me.

If you are just impatient to see the bottom line, then go ahead. You can always return to see the rationale behind the changes that I have incorporated into my STAR. None of the changes requires PCB copper modifications, only different components to be used. No cutting or drilling, and all reversible (provided that the copper stays on the board!).

To see what I found when looking at the Receive path, go here.

Then, having a functional solution on Receive, I tried the Transmit path, expecting that it would be less troublesome.  It wasn't, really.

Following the study of both the Receive and Transmit paths, we at long last have a set of modifications that can be incorporated into the 2nd IF paths to put things right. The changes fit in with my stated philosophy of making no copper changes to the PCB and not adding extra 'fix-it' boards, whilst greatly improving the Tx and Rx paths through the 2nd IF and removing any original design errors and quirks.

My modification philosophy

I always try to propose solutions that can be achieved unobtrusively and without daughter boards or copper modifications to the PCB. If there is no ready solution, it is probably best to keep quiet.

Schematic of modifications

Glenn VK3PE has kindly adapted the ComboStar circuit to show the Rx path  modifications.
Click on the image for a pdf file of the same thing.
Please refer to the 'Summary of changes' (below) for the latest values.
Mods schematic from Glenn

Summary of changes

The list of modifications I now have incorporated is:

Receive-only path:
Transmit-only path:
Common path to Rx & Tx:

Component Notes:

'PI' Pad alterations:

If the pot RV604 does not give enough signal at minimum resistance, preferably re-calculate the PI network values for less attenuation, or you can increase RZ3 (not RZ1) for higher level. Do not alter for higher level than necessary, which would worsen distortion and intermodulation in TR603. The output level pot RV604 should preferably be near (but not at) the minimum resistance end when the IF setup is complete. If it is at the end-stop, a feedback pole with C692 and op-amp internal resistors will probably cause instability; back RV604 off a fraction to avoid this.

That's it! Not too bad really.

Photographs of the IF2 modifications on my ComboSTAR

The photos below are of my ComboSTAR with the 'T' filter above incorporated. I have performed the IF calibration successfully on Receive, but have not yet been able to perform meaningful antenna sensitivity checks. It is no worse than before, anyway. The two 330uH inductors are in series in the place of RFC612 (680uH) and measure almost exactly 680uH. The nearer 330uH inductor is mounted on the pads vacated by C631. The PI pad now has 82-100-82R. All my receive side coupling capacitors (now only CZ3 and C633) are now 10uF ceramic X7R,  and I have modified the transmit coupling (C632) to be a 10uF and 47R in series in the first photo. The repair to a pad on IC604 from changing the chip (I soldered it mis-orientated) is clearly visible. I subsequently made a 'both sides in parallel' desolderer! The Tx gain trim RX is not yet set.
Fully modified 2nd IF

close up of the extra L in the C631 position

Caveats

Since I am not party to any of the original circuit design or modifications in the 2nd IF, it is possible that some of the things about which I comment are actually deliberate ways of tailoring performance. For example, the LF rolloff at 15kHz may be a deliberate way of making the response look more like that of a tuned circuit, with a peak 'at' the 2nd IF (15kHz). On reading the description accompanying the original design from which the circuit was cloned (DSP-10) it is apparent that this is not so; it is just a mistake. In the 1999 QST article by Bob W7PUA he says "No active supply decoupling is needed because the lowest frequency amplified (set by C32 [which is our C633]) is a few kilohertz", but it is actually above the 2nd IF.  It is my personal opinion that the circuitry in this area has gone a bit wild; it works but not as we know it (to paraphrase from Star Trek).  If you disagree with or dispute what I have said & done, please contact me and put me right.

My own results

In the next weeks I aim to complete fully my own STAR incorporating all these changes and any others that seem necessary. I have already modified the DDS filter & buffer as previously described. In the fullness of time I will let you know the results, which will mainly be 'works' or 'doesn't work' since I will not have "before and after" performance measurements. For me it will suffice that several peculiarities have been dealt with.

So far I have performed the full scope of modifications that are described above and verified the Receive path.
The IF calibration BASIC programme has no issues with the modifications.
I have no comparison data to see if the overall performance has changed, in terms of MDS or IP3.  Maybe someone can do this someday, but for me it is sufficient to have exorcised the 2nd IF stages of so many wrongs.  My personal 'take' is that mismatches and filter errors in any part of STAR, especially those which are so easily remedied, are simply not welcome. Correcting them will help me to be happy that things are not just 'working', but 'working properly'.

Has all this done any good?

There are loads and loads of changes to almost every part of the 2nd IF circuit.
These changes have now been incorporated into an original PICASTAR (modular) and a ComboSTAR.
Measurements have been made by an independent Amateur of MDS on each unit 'before and after', carefully averaging readings and performing them conscientiously, which clearly show:

Modular STAR

Measured on 20 metres in a 3 kHz BW using Hewlett Packard test equipment,
MDS improves by 1dB, from -127 dBm to -128 dBm, in 'Best NF'
MDS improves by 1dB, from -122 dBm to -123 dBm, in 'Best IP3'
(I apologise to the kind Amateur who conducted these measurements for my careless error in presentation of his 'Best IP3' figure, and thanks to the reader who pointed it out).

ComboSTAR

Measured on 20 metres in a 3 kHz BW using Hewlett Packard test equipment,
MDS improves by 1dB,  from -128 to -129 dBm, in 'Best NF' mode.
This result was clearly a comfortable 1dB; it came with the comment "I was seeing less 'wobble' on the meter needle and it looked like nearer 1.5 but I'm still not going to add any decimal points to my quoted result".

This is probably the easiest 1dB improvement you can get, since it comes free of charge by correcting the original design errors. It may not be earth-shattering, but since it comes as a result of correcting the design it is nevertheless welcomed! 
I have received a comment that a mere 1dB improvement is not significant, but someone has also said that if it had become 1dB worse rather than better then it would have been heavily criticised...

At least it shows that putting things right in the 2nd IF does not make it worse! I am very happy with this result, albeit not earth-shattering - after all, this series of corrections was never done with MDS improvements in mind, but to remove unsound design elements. 

Are all these changes necessary?

Only if you wish to remove a host of errors from the circuit.

We all know that it is difficult to obtain a perfect design.
The silly thing is that by performing proper calculations and using a bit of ingenuity, it is possible not only to remove all of these problems, but to do so within the existing physical framework. Many of the original components are retained, often only small corrections being needed in order to ensure that the proper operation results.
Like me, you will be surprised at how many problems there are in this relatively straightforward area. You may decide that I am wrong; they are features of a good design. That is your prerogative. I have provided full explanations, plots, analyses, measurements and even told you what to do in order to remove these design errors from your STAR. It is entirely your choice if you do any or all of it.
All the changes I have shown are needed in order to purge the design of errors, some serious.
But you choose.

These mods provide:

Should you incorporate the changes?

Existing STAR users would be wasting their time. They have a pretty good transceiver. Access is probably difficult. The changes won't give earth-shattering improvements in Rx sensitivity and will very likely not be discernable.

New constructors, who have easy access to the components and who are prepared to use the new values should decide for themselves what they wish to do. Remember - these are still Unofficial Modifications.
If you ask my opinion, I would say that sanitizing the couplings and matchings, together with removing distortion, is beneficial! The 2nd IF will be working as you expected it to. You decide, I have provided all the information to the best of my knowledge and ability. Compromising the signal just for the sake of it seems strange to me, especially when it is easy to get it right.
Let me know what you think, especially if you decide to go ahead! As always, I can offer no guarantees of anything.

Practical Tests

OK, my approach has initially been theoretical, with calculations and simulations. So what do things really look like?
For me, it was not easy to perform measurements, especially on the 'before' which no longer exists on my ComboSTAR.  Fortunately, others have leapt in and bridged the gap.
Both Roderick VK3YC and Glenn VK3PE have built an N2PK VNA; Glenn discovered that these will operate down to about 10kHz (the internal transformers seriously distort waveforms below this). This is good enough to plot the actual behaviour of chunks of the circuit. Both Glenn and Roderick have kindly lashed up the original and modified filter sections and measured their responses.

VK3YC setup of original filter on his VNCVNA Measurements comparing original & new diplexers

Luckily Glenn is constructing yet another STAR (!) and has measured the whole 2nd IF response with the original components, then has adapted it and re-measured with the revised parts.
See the actual results. 


Please contact me with your thoughts.

Credits

Thanks to the following people, without whom this would probably have fallen by the wayside: