PICASTAR Bidirectional Amplifier

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 Bidirectional Amplifier, and 'carry on regardless', it is perfectly apparent to me that if the original design team had been aware of any 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 problems 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.

The Bidirectional Amplifier

The Bidirectional Amplifier is used as a buffer amplifier between the Roofing Filter and AD603 IF Amplifier (on Receive) and between the 2nd Mixer and Roofing Filter (on Transmit).  A cunning circuit arrangement using diode gating changes the direction in which the signal flows between Tx and Rx, hence the name Bidirectional Amplifier.

The circuit first appeared in the "Belthorn 4" some years before, and like so many other areas of STAR, seems to have been 'cloned' without any further investigation of suitability.

This amplifier uses a single J310 FET to provide amplification and some input-output isolation. At the Roofer port, a transformer wound on an iron dust core T37-2 acts both as a low-Q resonant load and a conjugate match to the impedance of the crystal filter (900R//10pF or 910R//25pF, depending on the filter used).

At the AD603 IF Amp / Mixer port, the amplifier is coupled using an untuned ferrite transformer of 10t:3t wound on an FT37-43 toroidal core.
Both ports are intended to show 560R, defined by resistors of this value. This is borne out by the ferrite transformer, whose ratio transforms the 50R mixer output by 102 / 32, 100/9 = 555.6R, very close to 560R.

This does not work out properly for the 2nd IF amplifier, IC 602 (AD603) because the input of the AD603 is a closely controlled (laser-trimmed) 100R, not 50R, and no correction is performed.  This means that the bidirectional amplifier is mis-terminated by the AD603 IF amplifier when operating in Receive; this needs consideration since it must affect the bidirectional amplifier performance during Receive. I deal with this later.

[add a schematic scrap here]

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.

Design Considerations

With any amplifier of this type, there are many parasitic elements that can apply.
We are used to such things as the input and feedback capacitance of the FET, but here we also have the corresponding capacitances introduced by the 'off' diodes used for direction switching, together with several normal components that affect impedances, such as the resistors used to bias the diode switches.
Because the amplifier is designed to operate internally at 560R, the effect of parasitic capacitances is much larger than would be the case for a 50R amplifier, as you would expect.

For example, when switched to 'Receive' by means of appropriate voltage levels on '+12VRX' and '+12VTX', diodes D607 and D609 conduct, allowing them to feed the gate and drain respectively. At the same time, D608 and D610 are reverse-biased. These are silicon band-switch diodes, BA244, although many constructors will have used 1N914-type diodes.  BA244 diodes have a reverse capacitance of around 1pF; 1N914 around 4pF. Although this doesn't sound a lot, it represents an unwanted impedance of 15k or 3.7k at 10.7MHz, and has a significant effect especially for that diode which couples back into the input (and hence is effectively in parallel with the FET feedback capacitance).

In addition, the input impedance is reduced directly by the presence of the 4k7 resistors R618 and R619, which serve to feed the switching voltage to the input-switching diodes. These two resistors, together with the 22k gate resistor R634, inevitably appear in parallel with the input circuit, by capacitive coupling for one of them.
The net value of these in parallel is 2.12k, which effectively appears in parallel with the 560R matching resistor R605 or R606 depending on direction, reducing the value to 443R which is clearly not right for a match.
To regain the expected 560R, the value of the resistors R605 and R606 can be increased so that when shunted with 2.12k the correct value appears. Using 750R gives a result of 554R, which is just right.
This value also shows correctly in circuit simulations, and is also that which was suggested by Roderick, VK3YC in the homebrew-radios Yahoo group.

Having increased the 560R resistors to suit the fixed shunting effects on the gate, it is now necessary to do something to restore the correct value to the drain circuit, since 750R is clearly not correct.
The obvious shunt value to apply is the same as the parasitic shunting on the gate, 2.12k. Roderick VK3YC suggested using a 2k2 on the drain, in series with a 100nF to ground in order to block the d.c. path.
In simulation I have found that 4k7 offers a better match when all the capacitive parasitics are considered. With 4k7 there is no real need for a d.c blocking capacitor, since at 12V the extra current is small (<3mA) and dissipation not exessive even for a surface-mount component.

Please be aware that in the calculations on this page where bus-switch bias resistor networks are considered, the values I have used are those which I recommend for proper biassing. In this IF area the lower 10k resistor is either replaced by 3k6 or shunted with 5k6. Shunting 10k with 5k6 gives 3k59.

Diode Types

Earlier we saw that the gating diodes have quite significant reverse capacitance. There is always one of these diodes whose reverse capacitance connects between output (drain) and input (gate) of the amplifier FET. The effect of this capacitance is amplified by the stage voltage gain (Miller effect), as is the inherent feedback capacitance of the FET. Had the amplifier used a cascode connection, the FET feedback capacitance at least could have been isolated. Unfortunately, there is not really sufficient headroom even for a bipolar 'top section' to be used with the J310, so this approach was discarded.

The preference then is to use very low-capacitance diodes. The best I have found are PIN diodes, with off-capacitance of 0.5pF, such as the BA482 which is still available occasionally. Fortuitously I used four of these when I built my STAR since I could not get the specified BA244 (which is a low-capacitance conventional diode) and I had several BA482 in the spares box.
If PINs are not available, or the expense is unattractive, and BA244 cannot be obtained, then I suggest using pairs of 1N914 (1N4148) diodes in series in each diode position, which halves the otherwise large 'off' capacitance.

FET Biassing

Well, actually the FET is NOT biassed. The gate is at the same potential as the source, since there is neither a negative supply to the gate nor is a source-resistor used. This highly unconventional design means that the FET runs outide the manufacturers' figures. With Vgs=0V, the drain current is whatever the device characteristic IDSS
makes it.
In addition, being a JFET, the gate is actually a diode juction with the channel (s-d), and is expected to be biassed -ve of the most negative part of the channel (the source) in order to avoid gate leakage, or worse still, forward conduction of the diode. Because of the zero bias, the gate will conduct as a diode into the source when the gate voltage approaches 0.6V, just like any other silicon diode. If this happens, serious non-linearity must occur! This is categorically not a good way of using the FET.  It is made more likely to happen by the voltage step-up inherent in the impedance transformation.
Although it is easy to see this clipping when testing in the Tx dirction with a steady signal, it will also happen in the same way for high amplitude received signals within the roofer passband. Because the impedance of the base circuit is 560R, the onset of gate clipping corresponds to a much lower level in the 50R circuit. At the 50R input to the roofer, ignoring filter losses, a peak of 200mV (-4dBm) will cause gate-clipping.

Below is the actual gate waveform on transmit with 'RX' at zero ohms. The gate clipping is severe and is seen as the flattened positive peaks of the waveform at 0.5V or 0.6V:
Intrinsic Gate diode clips Tx waveform at 0.6V
This gate clipping is completely avoided by the proper biassing with a 47R source resistor.

If we study the FET data-sheet, it is apparent that the best dynamic range is obtained if the device is biassed to a drain current of 20mA. This is easily obtained by inserting a resistor of 47R into the source lead, which provides both conventional auto-bias and a measure of stabilisation of operating point by virtue of the d.c. -ve feedback.
We have the choice of bypassing this resistor with a capacitor in order to give the highest gain, or leaving it unbypassed in order that the stage can benefit from the a.c. negative feedback also. I have chosen not to bypass this resistor; the stage gain is reduced somewhat, but this actually has benefits:
By correcting the bias and reducing the gain by negative feedback (source degeneration), the overload on Transmit is prevented and non-linearity on Receive is reduced greatly.
The gain on Transmit is still more than enough, and on Receive any loss of gain is readily compensated by altering the PI-pad in the second IF path. In fact, lower gain in the receive path improves the overload capacity (IP3) of the whole path and also improves linearity of the 2nd mixer, so reducing gain in the bidirectional amplifier is generally beneficial, as well as the negaitive feedback improving linearity of the bidirectional amplifier itself.

By adding this simple resistor the stage becomes properly biassed and will not only handle much larger signals without distortion, but also works 'as the manufacturer intended', so that the operating parameters actually appear on the device data-sheet!

Simulation

Below is the simulation schematic and a plot of input & output signals. The input impedance is frequency-dependent, but you can see that at 10.7MHz the signal drops by almost exactly 6dB from the generator (50R source) level, so it is 50R at the IF.
Simulation schematic
Plot of above schematic

The broad peak in output at 10.7MHz is from the roofer matching circuit T601. In practice the trimmer VC604 will be adjusted to get either the best amplitude or, better, to get the flattest filter response. Note that when simulated in the Receive direction the tuning shifts slightly. Either tune for the direction you feel needs to be best, or a setting between the best for each. It is probably easiest to tune on Tx, since a single-tone signal can be produced easily by keying the Tx, but remember not to tune with a 'scope probe on the filter pin; most x10 probes are between 15pF and 30pF and will result in a considerable shift when removed! It is best to 'scope on the other side of the filter, preferably at the 50R point on the side of L601away from the roofer, which will then not affect tuning VC604.
The gain shown is in dB, but since the impedances differ largely between sides (50R input, 910R output) this is not a power gain but purely derived from voltage gain. Bear in mind also to allow for the 6dB source loss, i.e. the peak gain relative to the transformer input is 6dB higher than the raw generator.

Changing the 750R values to either higher or lower values worsens performance, so it seems that 750R is correct.
 
Note that the simulation does not include the imperfections of the transformer T602, which are covered next.

Transformers

The final element worthy of attention is the input & output coupling transformers.

In the fullness of time I hope to present further measurements of the cores based on VNA measurements that should show losses and resonant effects more clearly. For the moment I can only 'show it like it is' for the transformers in actual use in the bidirectional amplifier as a whole.

Roofer coupling

In the fully modified STAR, the original plain transformer at the crystal filter is replaced by a clever low-Q resonant transformer (T601) that is a cross between a normal 'L' conjugate match circuit as used at the other side of the roofer (L601) and an auto-transformer at the FET ports. The tuning is performed by the trimmer VC604, which allows both the roofer and bidirectional amplifier to see the correct match.  This tuning can be correct either on Tx or Rx, but is unlikely to be right for both since the topology of the amplifier varies and so does the capacitance presented to the transformer.  Although I have sought ways of normalising this, they are not really satisfactory and compromise the operation of the amplifier, but I believe that due to the low Q the mis-tuning effect is not severe.
In general, this is a cleverly implemented element and should not be changed. The resonant nature absorbs the amplifier parasitic capacitances into the tuning element, which is ideal even though it varies slightly (by about 8pF) between Tx and Rx.

IF/mixer coupling

This is a conventional ferrite transformer of 10t:3t, to match the 560R amplifier impedance to the 50R outside environment of the 2nd mixer or AD603 IF amplifier. The AD603 amplifier actually presents 100R rather than 50R, which is another design oversight, but for now let us consider it to have been corrected to 50R, maybe by a 100R shunt being added. If this is not done, the bidirectional amplifier output will be unmatched in Rx, which might adversely affect it's operation.
 
Roderick VK3YC, Glenn VK3PE and I (they both have VNAs <g>) are looking into the transformer design. A plain ferrite toroid does not have high coupling factor, which can upset matching. Attempts at using a BN43-2402 binocular transformer give vastly improved coupling, but the type 43 material is not excellent at 10MHz and also the 10t winding resonates at below 10MHz with the tiny (2pF or so) winding self-capacitance, upsetting impedances.
We are still investigating this with several transformer arrangements, including:
The binocular cores give far better coupling than toroids, even with the toroid primary wound over the secondary, but the inherently higher winding self-capacitance may defeat this by resonating with the winding inductance. Getting this transformer to be as good as possible is good policy, especially bearing in mind the 560R impedance at which it must work.

Ferrite Notes

Whilst it is often expected that ferrites are 'wonder' materials, they have a downside. High permeability ferrites, such as type 43 material, provide very high Al values at low frequencies, but this falls rapidly as frequency increases (over 1MHz in the case of type 43). They also have rather high loss, so are not ideal for conventional flux-coupled transformers, where the windings couple by virtue of magnetic field in the core. Ferrites such as type 61 material have lower initial permeability, but this remains level up to much higher frequencies, with much lower core losses also. At our IF of 10.7MHz, the apparently beneficially high permeability of type 43 material has fallen to a value close to type 61 material - it also has very much higher losses.
The plot shows the comparison between types 43 and 61 ferrite using a single full turn on BNxx-202 cores, and has been produced by Roderick VK3YC.
BN43-202 vs BN61-202, both with one turn
The above plot shows clearly the much higher Rs (loss) in the BN43 core at 10.7MHz, together with the inductance for the type 43 material falling rapidly above 2MHz (the dip at the start might be due to the VNA measurement). The initial inductance advantage of 2.07uH vs 0.464uH rapidly dwindles, with the type 61 holding up whilst the type 43 actually gives less inductance at frequencies above 28MHz!

At 10.7MHz the equivalent series resistance Rs of the type 43 winding is already over 75R, whilst the type 61 material Rs is just above zero ohms, much less lossy.
It is also very clear that calculations should bear in mind the actual permeability of the core material at the frequency concerned! Even in my own simulations I have been guilty of using the headline figure in order to calculate the winding inductance; this applies at low frequency (for material 43) but at the 10.7MHz IF is completely wrong. The ratios for the two windings will be correct, but in the simulation I have also not made any allowance for the considerable core losses that can be viewed as a shunt resistor, reducing the overall impedance.

The plots show clearly that type 43 material is not always the best to use. In fact, I have concluded that it is often the worst due to the atrociously high losses! It is good for interference suppression...

A normal transformer, with separate primary and secondary windings, is flux-coupled, since it relies on the magnetic material to provide the coupling between windings. If the core material is lossy, then the transformer will be inefficient. If, however, the transformer is wound with multifilar windings, then the coupling at higher frequencies is by transmission-line coupling. In this case, the flux-coupling due to the core happens only at low frequencies; at higher frequencies it occurs by the close proximity and transmission-line effect of the wires.  It is, therefore, always worth using multifilar winding, to minimise core effects at high frequencies, but not always possible.
In the case of T602, it is simply not possible to use multifilar windings without redesigning the circuit, since the ratio of 10:3t cannot be achieved by series connection of multifilar wires unless enormous numbers of turns are used, with the consequent high winding capacitance giving a fresh set of problems, or by using 'one turn' of 13-filar wire, which is ridiculous. A ratio of 9:3t rather than 10:3t could be achieved with a quadrifilar winding, by connecting three 3t windings in series to form the 9t, but the impedance at the FET would then reduce and the whole circuit would need to be re-calculated.
Since T602 will remain as 10:3t ratio, it must have individual windings and will be a flux-coupled transformer. The core material is therefore important, and it is apparent that type 43 material at the narrow operating frequency around 10.7MHz provides not only rather low inductance but also atrocious losses, which affect the impedance seen through the transformer as well as simply losing signal. Type 61 material is very much less lossy and still has a respectable Al, so is the material that I believe is right for this transformer.

Transformer tests

I have wound a number of different types of transformer on different core materials, in order to attempt to assess the suitability of each. The results for each transformer, used with the fully modified amplifier, are:
These are for 'regular' 10:3t transformers (not 10:3+1) wound on various
cores, with Vpp voltage reading, together with a reference using 51R on
the T602 pins 2-4 (3t) with no transformer.
All readings are Vpp with the transformer driven by the 2nd mixer in Tx mode whilst keyed:

Ref
 Core type
 Gate
 Drain
 C665 (input)
 X
1
 50R 1% resistor
 --
 --
 0.70
 --
2
 BN43-2402
2.0
 7.0
 0.72
 2.4
3
 BN61-202
 2.0
 7.0
 0.68
 2.3
4
 FT37-43
 1.7
 6.0
 0.55
 1.83
5
 FT37-61
 2.4
 7.6
 0.9
 3
6
 BN43-2402 'half-turns'
 1.55
 5.55
 0.6
 2.0
7
 BN61-202 'half-turns'
 2.3
 7.6
 0.69
 2.3
8
 BN61-202 'half-turns' with extra 1/2 to feed AD603
 2.3
 7.7
 0.7
 2.3
9
 BN61-2402
 1.9
 6.9
 0.64
 2.13
10
 BN61-2402 'half-turns'
 2.4
 8.0
 0.78
 2.6

Ref 6,7: 'half-turns' means that each turn is a single pass through only one hole of the binocular core. Conventionally one turn passes through both holes, 'there & back'. Initially this was merely experimental, since the '3t' winding gives an unbalanced number of turns in each hole.

Column 'X' is the 'G'ate voltage that would be obtained from a perfect  10:3 transformer fed with the 'C665' input voltage, i.e. 'X' is 10/3 * 'C665'. Note the disparities except for (7)!

These are all measured as best I can in Vpp with the 'scope, so are subject to the accuracy & resolution of that. This will probably be why the FET gain apparently varies between measurements 5) & 7) etc.

The graph shows these results in what might be a clearer manner. For each type, the 'key' parameters are the height of the yellow bar compared with that of the reference 50R (this shows how correct Zin is) and the ratio of the height of the blue & green bars for each transformer, which shows how accurately the transformer secondary voltage (at the FET gate) reflects the winding ratio of 10:3.
Graphical representation of transformer measurements


My criteria for choosing 'the best' core are, in order of choice:
  1. The primary (3t) winding must present the same load as a 50R resistor, as determined by the voltage measured at that point. We cannot be sure that the signal source (2nd mixer on Tx) is itself 50R, but if our circuit gives the same level as a 50R resistro than our circuit is presenting 50R impedance,
  2. The secondary must show the primary voltage stepped up by the winding ratio (10/3), to show that it behaves with good coupling and lack of resonances. Note that originally I intended measuring at the Drain in order to avoid the x10 probe capacitive loading on the Gate, but this introduced another error source (the gain & distortion of the FET) and is not easily calculated from the winding ratio.
If we assume that the 50R is the 'proper' loading that we aspire to achieve, then the FT Toroidal transformer results stand out as being furthest away from proper matching and the binoculars give good results. This is probably due to the tighter coupling inherent in the binocular cores.
The 'half-turn' BN43-2402 H is not any good, but the half-turn BN61-202 H is best of all combinations, showing an input match close to 50R and accurately transformed signal on the secondary.

Interesting is that the 'full-turn' binoculars are so similar in spite of the different core materials. The larger BN61-202 has considerably more wire, so the winding capacitance will be higher, but because it is less lossy and has good Al at 10MHz in spite of apparently being a lower Al material, it does well. The small BN61-2402 is surprisingly not awfully good - even the 'half-turns' version is nowhere near as good as the larger BN61-202 core.

I judge this (half-turns BN61-202) to be easily the best transformer. The 'half-turns' BN61-2402 is probably second best, followed by the other binoculars which are not so good.

The two toroids are not good at matching the 3t impedance. I checked the turns on both toroids, since the results are so disparate, and they are correct.
The FT37-61 shows interesting results, with high gate (secondary) voltage but an input impedance of well over 50R and hence a higher input voltage. I don't pretend to understand this - maybe it is near self-resonance at 10.7MHz? For a 10:3 ratio and 0.9 primary voltage, the secondary should be at 3.0V if it were 100% coupled, so I suppose it has high leakage inductance, which seems to be a problem with flux-coupled toroidal transformers, together with the high core losses of type 43 material.

My choice is the half-turns BN61-202, which is head-and-shoulders better than the others. This is in spite of my personal uncertainty about how this 'odd turns' transformer actually operates - it is simply best.
In second place is the half-turns BN61-2402, then the full-turns BN43-2402, BN61-2402 and BN61-202 are not atrociously far out, with the two toroids coming next and the 'half-turns' BN43-2402 being worst, probably due to inadequate winding inductance together with high core losses.

It is not worth using anything but the 'half-turns' BN61-202 unless this core is simply not available, in which case the BN61-2402 'half-turns' or if desparate the BN43-2402 could be used.

The effect of an extra turn for better matching to the AD603 input is covered below

Schematic of changes

The schematic of the revised amplifier using BN43-2402 rather than BN61-202 is below; click on it for a pdf copy which does not yet include changes around the AD603 input & output.
Schematic of Changes to Bidirectional Amplifier
Rather than buying 750R resistors for R605 & R606, use two 1k5 in parallel for each - easy with surface-mount parts.

Here is what the changes look like (this is Glenn's ComboStar; he used all surface-mount parts for the mod).
Glenn's mods to the bidirectional amp.
You can see the 47R emitter resistor between the pad and the FET body, and the 4k7 drain load between the same 0V pad and the diode junction. The 750R values are made up from pairs of 1k5s.
The binocular T602 is a BN43-2402, since Glenn has no BN61-2402 or BN61-202 available. He is happy with the performance but intends incorporating a new transformer when it is available.

AD603 Mismatch

Earlier I mentioned that the AD603 laser-trimmed 100R input resistance is not a proper termination for the bidirectional amplifier when in 'receive' mode, so I will deal with this here as a function of the bidirectional amp changes.

Whilst the 2nd Mixer is designed as 50R, which is fine on Tx, the AD603 input impedance is laser-trimmed to 100R +-3% and is therefore a considerable mismatch for the bidirectional amplifier on Rx. The AD603 doesn't mind being fed from 50R, but the bidirectional amp is designed to operate into a load of 50R, not 100R. This must be corrected in order for the bidirectional amplifier to function as designed!

We could add a 100R shunt (corrected for the circuit parasitics) to produce a 50R load for the bidirectional amp; this is as shown in the AD603 data-sheet, but this would throw away half the signal.

Roderick VK3YC has suggested winding an additional 1t (ideally it would be 1.2t, but of course this is impossible) on the '3t' side of T602, which brings that point up to 560 / 10^2 * 4^2 = 89.6R, then isolate IC603 pin 9 either by track-cut or lifting the pin so that the 89.6R can feed to the AD603, whilst the original 3t 50R feeds to the mixer2.
Either the extra turn can be connected between pins 11 & 9 to obtain bias, carefully observing phasing, or the bias arrangement can be modified by moving C665 into the earthy side of the transformer, so that a 'tapped' winding can be used. These are equivalent, but moving the C665 makes correct phasing easier since the 1t is an overwind on the transformer rather than a separately connected winding. Since the transformer has flying wires, the latter repositioning of C665 is not difficult and is the preferred method. It is not the way I did it myself, though...

The 100R input of the AD603 now doesn't quite match to the winding of 89.6R. The mismatch is improved by the (modified) bias network (R623//R624, 10k//3k6 = 2.65k) in shunt, but is worsened by the bus switch on-resistance of 4R (Fairchild) to 6R (TI) (say 5R) in series. In order to get a 'perfect' match we must put a shunt at the AD603 input.

The desired impedance is 89R, and we have 2.65k in shunt already, then 5R in series.

For 89R including the 2.65k in shunt in parallel, we then need another 92R to appear. This is made up from the 5R switch, plus the parallel combination of the AD603 100R and our new shunt.

Thus the AD603 // new_shunt must equal 92R - 5R = 87R.
To get 87R with a // combination with 100R needs a shunt of 680R (100//680=87.18R).
Hopefully we can put this between AD603 pins 3 & 4 with no d.c. block,else we need a series 100nF since there is voltage on both sides of C643.

From the AD603 data it seems as if no d.c. block is needed if the 680R is placed directly between pins 3 & 4.

Note: Xc of C643 (10nF) at 10.7MHz is 1.5R which is not accounted for; I suggest changing it to 100nF as used elsewhere in this type of function, but this is optional since the impedance error is small.

Extra turn - effect during Tx

In order to assess the effect of this extra turn, which is parasitic on Tx, I added the turn to the 'half-turns' BN61-202 transformer, connected it to the isolated IC603 pin 9 and to the bias network R623, R624 in the correct winding phase, and re-measured the a.c. voltages.
These were unchanged. In addition, there is now a voltage at the pin 9 of the bus switch (which is open on Transmit), which measures as 0.95Vpp. The calculated value for a 4:3 stepup of the input is 0.92Vpp, so the measurement is well within my measurement accuracy.
Of course, the measurements during Rx will depend on the received signal amplitude.

AD603 Output Match

Now we can consider the AD603 output matching. Right now the low Zout of the AD603 feeds the 2nd mixer through a network of 100R series and 68R shunt nominally, which looks to the mixer like 40.5R rather than 50R. Why?

The output impedance of AD603 at <10MHz is typically 2R. Assuming that this figure applies also at 10.7MHz, then it is effectively a voltage source and we need not take output impedance into consideration (can consider it zero).

If the 68R RZ1 (a mod) is increased from 68R to 100R, then the feed will be pretty close to 50R. It is affected by the bus-switch bias network R621//R622 (2.65k) and the series '4-to-6R' of switch resistance, just as is the input side. To match 50R with 5R of switch resistance we need to offer 45R. With RZ1 at 100R we have 100//100//2k65 = 49R. To get 45R we need RZ1 to be 91R (gives 46.8) or 82R (gives 44.3) of which 82R is better for TI switches and 91R for Fairchild! Or just use 82R and be quite close. Although it is apparent that the bus-switch on-resistance has been neglected generally in the STAR design (it has been assumed to be zero), I cannot see any reason for not considering it, especially since it is easy to do so.
So RZ1 changes from 68R to 82R.

In order to pad down the 100R input of the AD603 we must add a 680R shunt resistor at the device input pins. It is easy to fit an 0805 resistor on the bottom of the board (opposite side to the IC) between the pin 3 & 4 pads of the AD603, whether or not the SMT version is fitted.

Just one final check, to ensure we have not loaded the AD603 output beyond spec:
The load from RZ1 and mixer+switch is 55//82R = 49.1R
There is also a series resistor R602 (100R) giving a load of 149.1R. This does not include the gain resistor R610, which could be included as part of the load.
AD603 sees 149.1R // R610 (2k2) = 139.6R which exceeds the minimum 100R load spec. and is therefore OK.

So what did M0RJD do?

I decided that since the BN61-202 'half-turns' arrangement gives the best result in the practical measurements, it is the best to use. This, I believe, will still apply even if it later shows to be due to a coincidence of resonances, or whatever. It behaves as a 'perfect transformer' in these measurements.  I used fairly substantial wire (around 0.5mm dia, from a cast-off mains transformer) and fitted this transformer vertically in the T602 position as shown in the photo, with a further 'half-turn' of PTFE insulated wire to feed the AD603 at higher impedance as described above. The wire is sufficiently robust to hold this rather large core safely.  I have also incorporated all the modifications described in this page and am happy with the performance. There is no overload of the circuit even with the full output from the DSP on Tx (i.e. with the resistor 'RX' at zero ohms) and I believe that all the mismatch problems and odd biassing arrangements are solved.

Your results may differ from mine, although I hope this will not be so. Please let me know if you use these suggestions and what is the outcome.

NOTE that it is very wise to compare the voltage to ground of each of IC604 pins 9 & 11 in turn. Pin 9 should have a signal 33% larger in voltage than pin 11 if the 'extra' turn is phased correctly. If the amplitude is lower, then the ends of the 'extra' turn need swapping over; check again afterwards.

M0RJD Bidirectional amp modified.
Above is Bob M0RJD's modified bidirectional amplifier, showing the BN61-202 transformer and the extra matching 'turn' for the feed to the AD603 in pink PTFE insulated wire. The track from the FST3125 pin 9 is cut at the point where it joins the track from pin 11.

Below is the modification as done by Steve G7WAS.
Steve's resistor mods
Steve's transformer

Summary of changes to the Bidirectional Amplifier and AD603 IF amp circuit

Please verify this is correct & let me know of any errors.

R605    Replace with 750R (e.g. two of 1k5 in parallel)
R606    Replace with 750R (e.g. two of 1k5 in parallel)
R624    Shunt with 5k6 as part of the bus switch bias change
R622    Shunt with 5k6 as part of the bus switch bias change
RZ1      Replace with 82R
Add      4k7 from TR601 Drain to ground
Add      47R in series with TR610 Source
Add      680R between pins 3&4 of IC602 (easiest 0805 on underside, directly between the DIL pins)
T602     Wind a Binocular core of your choice as described above. Best (I think) is BN61-202 with 0.5+1.5:5t as described.
IC603    Isolate or lift IC603 pin 9 to take the extra 'half-turn' - see text and photos above for connection & phasing.