STAR LO Investigation

In 2015, it has become apparent that constructors can and do end up with too little or a very distorted LO signal to the Magic Roundabout on 20m only.
This is due to the inclusion in the Matrix of D405, which energises the 2nd (bandpass) filter on 20m, as well as on higher bands.
This was not the case in Peter's Radcom articles as available on the Group, and wisely so.
Even the nominal filter is on the LF slope when the LO is set to 20m; with worst-case components it is significantly more so, and if the parts (likely the inductors) are even lower than expected at 40+MHz, as are many ferrite-cored inductors, then the wanted fundamental can be almost entirely suppressed, leaving mainly higher harmonics - not at all desirable.
Fortunately it is easily cured by cutting out D405 from the matrix. This will disable the bandpass filter on 20m and entirely avoid the problem. Because we all use fast DDS clocks these days together with the AD9951, the residual rubbish is minimised and is, anyway, likely to be far less trouble than having the filter present!
As always if you have a fully working Star then it would be pointless to open it up to make the change.
Thanks to Hank, WZ3HMT, for his work on this and for not suffering in silence.

Now, back to the LO stuff!
If you have no time for the background and maths, see the component values and schematic

Send comments or criticisms, please!


What constructors have said (good & bad!)

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

Many STAR constructors have suffered from low LO output, which results in the H-Mode mixer drive circuit (squarer) self-oscillating either during LO signal transitions or, at worst, continuously.

It is generally accepted that a level of +7dBm into the 50R input of the squarer  is needed for satisfactory operation. My LO, built to the published design on Glenn VK3PE's excellent ComboSTAR PCB, only produced +3dBm at the high frequency end of range (10m band), so I looked for solutions to the Yahoo! "homebrew-radios" discussion group.
It soon became apparent that there were a number of fixes in use, ranging from using a different LO buffer amplifier transistor type, to replacing the whole amplifier with an AD8008 design! Others had success with passive component changes.
On removing an amplifier input-matching resistor and shorting a decoupled emitter resistor, my STAR LO reached the magic +7dBm at 10m. On the journey to this I had investigated and analysed the DDS circuit, filter, amplifier and H-Mode squarer, changing and verifying every single component, and coming to the conclusion that nothing in my build was 'wrong'; it just didn't produce enough signal!
I suppose that, having achieved the +7dBm level, I could have accepted the result and moved on to other areas, but I am a curious person and felt that there should be a universal solution rather than individual 'fixes'.

The LO circuits are shown below as extracts from the PICASTAR documents Copyright of Peter, G3XJP and drawn by Glenn VK3PE.

The official filter and amplifier circuit remains the same as the original Pic 'N Mix design by Peter G3XJP, but the DDS area has changed over the years as better ideas and DDS chips arrived.

Background

It may be useful to see the calculations that establish the output level from the AD9951 DDS circuit. If you are a masochist, look here. If not, just believe that the transformer primary (6t c.t.) has 1Vpp across the six turns!

Output voltage calculations for the older DDS configurations, using AD9850 and AD9851, can be seen here.

Measurements

I have shown all a.c. measurements as Vpp, since this can be measured unintrusively with an oscilloscope and is independent of prevailing impedance. The use of a x10 frequency-compensated probe somewhat limits the lowest amplitude signal that can be seen on a 'scope, but we are trying to make big signals, so that limitation only applies when trying to measure beyond the filter cutoff.

DDS

The current DDS solution for all new constructors uses the Analog Devices AD9951 device, which has significant improvements over the previous parts used in STAR, the AD9850 and AD9851. AD9951 uses 1.8V and 3.3V internal logic as opposed to the earlier 5V internals, and is much more easily damaged electrically. The higher permitted input reference clock speed of up to 400MHz, together with the 14-bit DAC (10-bits on the AD9850/51) allow much better synthesis with reduced spurs.

VK3PE's ComboSTAR PCB has provision for direct assembly of the AD9951 and its peripheral circuit onto the PCB, but it can instead use the same DDS Carrier Board (a plugin sub-module) that is used on the earlier, Pic 'N Mix based DDS units. In theory this would permit the older, AD9850 or AD9851-based carriers to be used with ComboSTAR, but it is unlikely to be a winner due to the lesser performance of these devices. Most ComboSTAR builders will, like me, build the AD9951 circuit directly onto the main PCB, but some will use an existing AD9951-based sub-module.

Several people have questioned why the earlier LO had no trouble providing +7dBm, whilst the current AD9951-based LO struggles. Here is the reason:

"Older was Better"

The original 9850 in Pic 'N Mix was used single-ended. It fed a 200R 'source terminating' and d.c. load resistor, to ground because the 9850/9851 source current to a ground load rather than sinking current from AVdd (AD9951). The current waveform of 0 to +10.24mA into the 200R in parallel with the 200R filter produced 1.048Vpp of signal, with no transformer losses. Since the AD9850/51 have +1.5/-0.5V of output compliance, this is not a problem. This signal feeds the 200R filter, which is practically identical to that in the AD9850 data-sheet evaluation circuit (the first inductor value has been increased slightly to 1uH). This gives a reasonable signal level to the 2N3866 Buffer Amplifier.
The disadvantages of the AD9850 were the limit of 125MHz for the reference clock, making the sin(x)/x rolloff significant at high output frequencies, and the 10-bit DAC which limits the output step resolution. The whole 1Vpp output appears at the filter, with no transformer losses, but sin(x)/x rolloff still applies.
Before long Peter introduced the use of a centre-tapped transformer rather than directly coupled output. This not only gives higher output level but also better cancellation of even-harmonic distortion from the DDS.
Here things changed. The original single-ended output needed the 200R load resistor to complete a d.c. path for the DDS output current; the transformer-coupled version used the transformer primary to provide the d.c. path. As with the Analog Devices example circuit, the DDS output now uses no resistors, the a.c. load being the filter and subsequent termination, and the d.c. path being the transformer winding. No source termination means higher output signal, since no power is lost in source terminating resistors.
In fact, the output from this version is nominally 1.5Vpp into the filter.

Newer is better!

Then the super AD9951 arrived on the scene. The output circuit was provided by the Analog Devices Applications team after the STAR team had had little success at adapting the previous designs. Being a low voltage, small geometry device, the AD9951 cannot withstand any abuse such as overvolting the outputs. The previous transformer-only coupling could easily damage the device; with no load or with the increases in load impedance above 200R caused at higher frequencies by the filter characteristics, the device would exceed the +-0.5V output compliance limit and not like it.
The AD recommended 9951 circuit puts 100R directly across each half-primary of the output transformer. This limits the DDS output excursion to +-0.5V even with no filter connected (barring commutation spikes due to transformer leakage inductance) and produces 1Vpp across the full primary when loaded with 200R by the filter and amplifier.
output of 9951 DDS
This DDS naturally became the part of choice for new builds; it has far superior performance, and a replacement DDS carrier plugin was designed to give easy upgrade on existing Pic 'N Mix based STARs. With a 3+3t:6t transformer, the output is nominally the same 1Vpp that appears across the primary. This feeds into the DDS reconstruction filter as before.

Because of the re-introduction of source termination at the DDS with the AD9951 design, the output to the filter reduced from 1.5Vpp using AD9851 to 1Vpp, both nominal and subject to sin(x)/x reduction of amplitude at high frequencies. The net result of this is that the input to the amplifier is considerably lower than with the earlier transformer-coupled designs.

Amplifier

The single transistor amplifier circuit is shown below using the component references from ComboSTAR.
STAR original filter and amplifier

The Amplifier is a conventional single-transistor stage using a 2N3866. It is a clone of the design which originally appeared in QST of November 1981, Hayward & Lawson's "Progressive Communications Receiver", with identical topology and component values. It has a low input impedance at the base, and is transformer fed at the output to the 56R load on the MR.
The 2N3866 has a current-gain bandwidth product fT of 800MHz and minimum hFE of 25; at our top LO frequency of 40.5MHz the reduction of gain due to fT is dominant (at 40MHz the current gain becomes about 20 regardless of d.c. current gain). The input amplitude has already reduced from the 1Vpp at the DDS outputs by sin(x)/x rolloff, now we also have reducing current gain due to the 800MHz fT of the 2N3866. Oddly, this was offset somewhat in the AD9850/9851 transformer-fed implementation because of an imperfection in the filter - the input impedance rises well above 200R between 26MHz and 50MHz, peaking at about 330R at 31MHz and then reducing somewhat again. This removes some load from the DDS, whose output correspondingly increases and counteracts the reductions in amplitude. This is less pronounced with AD9951 because of the source termination, without which the DDS chip might be (and may have been) destroyed as the load impedance increases.

The Amplifier is fed from the filter via a series 180R resistor, R3311. This resistor pads the low input impedance of the amplifier input (about 20R) and acts as a good terminator for the Filter. Unfortunately, most of the signal is lost in this resistor. Taking the amplifier input Z as approximately 20R, shown by measurement and simulation, the potential division is large. If all of a 1Vpp DDS output reaches the output of the filter, the termination reduces it to only 0.1Vpp at the amplifier input. The amplifier has to work hard to achieve the required output of +7dBm into 50R, which is 1.42Vpp. Bearing in mind that we started off with 1Vpp (for AD9951), it seems odd now to amplify back up to pretty well the same voltage to overcome the loss in a matching terminator! Of course, we are also changing the impedance to 50R for the 'Magic Roundabout' (MR) squarer.

When using the AD9850/9851 DDS with transformer output, there was a good amount of drive available (around 1.5Vpp), so the Amplifier could relatively comfortably achieve over +7dBm output. With the AD9951 DDS, the lower level of 1Vpp, coupled with the small increase from filter input impedance rise and the loss from sin(x)/x loss, has left many constructors short of output from the amplifier. What can we do about it?

Conventionally the first recourse is to remove the 56R (R312 on ComboStar schematic), which is shunting some base drive from the amplifier for no good reason. The amplifier input impedance is only 20-30R itself, not high impedance, so the resistor is nugatory and serves only to reduce the gain.  Some people, me included, still did not achieve +7dBm. A cure for this has been proposed: to use a 2N5109 in place of the 2N3866. 2N5109 has an fT of 1.2GHz, 50% higher than 2N3866, which together with a higher minimum gain makes it able to provide more amplification.  This said, I know of one case where 2N3866 gave significantly better output than 2N5109!

But this is only tweaking things. It is quite disconcerting for new builders, just STARting out, to find that they need to adjust this, change that, and so on, in order to achieve the performance that seemed so easy for others. Maybe foolishly, I took up the challenge to find a solution.

The Solution

There may be other solutions; Robert G3WKU has suggested increasing the Amplifier input impedance so that less signal need be lost in the filter termination, and others have simply replaced the Amplifier with an IC version using AD8008.  G3WKU currently uses a 2:1 turns ratio transformer to match better the 200R filter output to the low input impedance of the amplifier and avoid the resistive loss in the present scheme.

Compromises

I have unilaterally decided that the optimal way for most STAR builders is one which does not require any extra components on the existing physical layout. Any revision should use straightforward components similar to the existing design and need only be compatible with use of the AD9951 DDS device.
Although this cannot yield results as good as a complete redesign of the LO (both electrically and physically), for most of us the need is simply to obtain good results using the existing PCB layout and with minimim disruption. For STAR builders who use the original, modular style of construction, it might no be so difficult to replace the whole DDS/Filter/Amp arrangement with something else, but ComboSTAR builders will probably prefer a less intrusive solution even if it does not produce the Best Possible Performance. Any practical design is a series of compromises; in this case the re-use of an existing physical layout.   As Voltaire once said, "The best is the enemy of the good". He probably wasn't thinking specifically about STAR, but you never know - he might have been.

The answer for us is to reduce the filter impedance from 200R.
This is almost trivial at the DDS end; it is only necessary to wind fewer secondary turns on the transformer (or unwind a few if an existing transformer has been constructed as suggested, with the secondary on top). At the Amplifier end the only changes, with luck, will be component values to reduce the gain!

As a later comment, I have become painfully aware of the poor performance at even medium frequencies of the type 43 ferrite material (see my comments relating to the bidirectional amplifier).
This material is very lossy at 10MHz, and the Al reduces between 1MHz and 10MHz until it is little different to the far less lossy type 61 ferrite. Since here the LO uses high-side injection, the LO frequency is always above the IF (10.7MHz normally) so there is great benefit in using a BN61-2402 core for the DDS transformer. Also, multi-filar winding causes a transformer to behave as a transmission-line rather than flux-coupled transformer at higher frequencies. This causes far less flux to be coupled by the ferrite, further reducing the effect of core losses. Anyone winding transformers for the LO would do well to use a BN61-2402 core with trifilar winding for the differential to single-ended trasnformation at the AD9951 output, and to ensure that the DDS Buffer amp output transformer is also wound bifilarly and preferably on a material 61 ferrite core. This will maintain good coupling and reduce losses, especially at high frequencies where it is so necessary.

Filter Impedance

Unfortunately low impedance filters have worse characteristics in general than high impedance filters, so obtaining the relatively sharp cutoff of the original filter is impossible if we are to have reasonable in-band characteristics (which we need).
Fortunately, we have opportunities now that were not possible with the original design; to stick with a problem that we can avoid just because of history is not a good idea. The AD9951 got us into this position; it will also allow a remedy.
Remember that the original AD9850 had a maximum clock input of 125MHz, putting the first image frequency fairly close to the wanted frequency at the higher frequencies and necessitating a sharp filter cutoff. The AD9951 will accept up to a 400MHz clock (sample rate) and the Butler oscillator will easily allow use of a crystal capable of providing in excess of 150MHz - the crystals are readily available and there is absolutely no reason not to use 150MHz or so, or even higher, especially on a new build.
Because of this, the first image frequency is much higher, leaving a larger frequency 'gap' into which we can fit the filter cut-off. At 150MHz clock the first image is at 150-41MHz, 109MHz, a 'gap' of 60MHz in which to roll-off the filter! Using even the highest permitted clock on the earlier AD9850 device (125MHz), the image is at 84MHz, leaving far less margin for tolerances and filter slope.
To make best use of this opportunity, we have to choose 'the right' filter impedance. There are a few things to consider, apart from whether we can actually implement a filter!


Initially I considered a filter of 139R. This is what you get with a 3+3:5t balun istead of 3+3:6t, one less secondary turn.
The filter designs quite easily with this impedance. Unfortunately the benefit from the reduced terminating resistor is rather small, also considering that the signal amplitude from the DDS has reduced. Instead of losing 90% of the signal voltage in the terminator; ending up with 0.1 x the filter output voltage at the Amplifier, we get 0.143 of the filter output voltage. But the DDS output voltage, and hence the filter output, have reduced because of the 5t rather than 6t secondary; from 1Vpp to 0.833Vpp.  The net result is that the voltage presented to the amplifier input changes from 0.1Vpp to 0.119Vpp, not an earth-shattering increase and not quite the answer.

I skipped the next step of removing 2 secondary turns, and decided to see whether it was possible to design a reasonable filter to suit the removal of 3 turns, making the transformer 3+3:3t. This would also be ideal for trifilar winding to increase coupling (reducing leakage inductance netween prinary & secondary), which should be of benefit, and results in the need for a filter of 50R impedance.

Having learned my lesson from the first attempt, I resolved to determine the benefit for amplifier input level before trying to design and construct a filter. The DDS output voltage is now halved, 0.5Vpp nominal. The padding resistor at the amplifier input needs to be around 30R to give a reasonable termination of 50R with the amplifier input impedance of approximately 20R. The signal arriving at the amplifier input is therefore 0.5Vpp x 20/50, which is 0.2Vpp, double what we had with 200R and hopefully more than enough to avoid low output problems!

Filter design

The filter design requirement was for an initial response that is level in amplitude up to 42MHz then falling off to -60dB somewhere in the gap before the first image frequency at 84MHz and hopefully remaining near or below -60dB thereafter, to achieve an improvement over the original filter for high images (the original filter rises to worse than -50dB on these images).

I used the excellent Elsie student edition analysis and design programme from James L. Tonne, without which the design and characterisation of the filter would be very tedious! I have taken the liberty of including some of Elsie's output here.

This is the new, 50R filter design for simulation and with normalised component values.
The new simulation circuit

The result of simulating the above circuit shows no enormous transmission disturbances within or outside the passband. The input impedance fluctuates within the passband, but so did the original design.
New filter simulation

For comparison, I have included the simulation of the original STAR 200R filter. This is the circuit
Original 200R filter

The simulation of the original STAR filter shows impedance ripple in the passband together with disappointing rejection of the higher images. The filter following the Buffer Amplifier further removes the images.
Simulation of original filter
Original STAR Filter, 200R


Having obtained a reasonable filter in spite of the low impedance, I altered the components to preferred values in the normal way, and checked that the inductors were reasonable to wind. The circuit and simulation above use these values, so they are representative.

An advantage of the final filter is that the capacitors are all of larger values than their counterparts in the 200R filter, so stray capacitance is less significant. The inductors are not the horrifyingly low value that I expected, so can easily be wound, and track lengths will not upset them. I then modified my STAR to use the new balun and filter components and checked the results.

With the original circuit, in order to achieve +7dBm, I had removed the 56R base shunt, R312 (ComboBoard reference) and completely shunted the emitter 5R6 (R316) in order to scrape +7dBm output at 40MHz in 'sig-gen' mode. I didn't change the 2N3866 to 2N5109; I didn't have a 2N5109 and this seemed quite uninviting to try, although it may have helped.

Results

After changing the balun and filter, without restoring the original amplifier circuit, the amplifier happily delivered over 2.25Vpp, in excess of +11dBm right up to the filter cutoff frequency of 45MHz, with +13.5dBm (3Vpp) at lower frequencies. The 3Vpp output is actually too high (!); it corresponds to a 6Vpp signal on the collector, but the static transistor c-e voltage is only about 7V so the transistor is in danger of clipping peaks if there is a slight amplitude increase for any reason.

With gain to spare, the 'problem' remaining is to reduce it in order to prevent the Amplifier overloading at any frequency! It's rather easier to lose gain than to -um- gain gain. Firstly I re-instated the emitter degeneration resistor that I had shorted before to increase the gain. The value actually 'jumped' to 10R, partly because of contemplating using a binary search for a good value, partly because my original build used two piggy-backed 10R in lieu of 5R6, afterwards further piggy-backed with a 0R to increase the gain - when I unsoldered the 0R, one of the 10Rs came with it...
This degeneration gives local -ve feedback in the circuit, in addition to the overall -ve feedback through the 1k resistor from output to base. The benefit is that variations in the circuit (the transistor itself, mainly) are levelled a lot, making the gain less dependent on the individual transistor. It also raises the input impedance, but this is mainly affected by the 1k overall feedback resistor.
Also possible and maybe desirable is to adopt W4ZCB's suggestion of reducing the capacitor on the emitter (10nF) thereby increasing the emitter-current feedback at low frequencies. This is worth bearing in mind if the output is Too High at lower frequencies (too high! That would be a treat!).  It may also be useful to reinstate the input shunt resistor R312 (was 56R) in order to pad down the input impedance of the amplifier if the emitter resistor is less decoupled; it will reduce the overall gain but make the amplifier input impedance and filter termination impedance less dependent on individual transistor characteristics. If you find these things necessary I would be pleased to hear from you.

The response with the emitter degeneration back in place, but with the resistor doubled to 10R is very promising. I measured the actual levels on my unit from 20MHz to above the high cut-off; because the second filter (following the Buffer Amp) stays active below its cutoff frequency in sig-gen mode (TrxAVR does this, probably Pic 'N Mix is the same), the results at and below 20MHz were not completely useful, but this is below the 20m band (14MHz + 10.7MHz IF is 24.7MHz). On lower bands the 2nd filter is switched out in normal use, so this is not a concern.

All signal levels are Vpp, measured on a 100MHz oscilloscope with a properly compensated 100MHz 'x10' probe.
R312 is omitted, see note in parts list.

Actual Frequency DDS-OUT of balun (50R)          

Filter out (50R)
Buffer Amp I/P (base) Buffer Amp O/P (50R)
Squarer I/P (50R) Output Level (dBm)
See Note
20MHz
0.26
0.54
0.3V
n/a
n/a
n/a
1
25MHz
0.38
0.6
0.45
2
1.5
10
2
30MHz
0.55
0.48
0.32
2.4
2.4
11.6
3
40MHz
0.35
0.35
0.18
2
1.8
10
4
45MHz
0.44
0.36
0.22
1.7
1.6
8.6 5
50MHz
0.45
0.11
0.05
1
0.05
3.98
5
55MHz
0.4
0.03
0.025
0.24
0
-8.42
6
60MHz
0.35
0.015
fuzz
0.07
0
-19.12
6
65MHz
0.32
fuzz
0
0.015
0
-32.8
6
70MHz
0.30
0
0
0
0
n/a
6

Notes:
1) In 'signal generator' mode, the secondary band-pass filter following the Amplifier is activated when normally it would not be at this LO frequency. The signal is below the filter pass-band, so the Buffer Amp is not properly terminated and no ouput reaches the Squarer Input.
2) At 25MHz, where the 2nd filter is beginning to cut response, the squarer input is 1.5Vpp, +7.5dBm, but the amplifier output is higher at 2Vpp, +10dBm. The 2nd filter may be presenting a load greater than 50R, which would cause this.
3) At 30MHz up to 40MHz, corresponding roughly to 20MHz to 30MHz at RF, the 2nd filter is in its normal passband and the output to the squarer is from 1.8Vpp to 2.4Vpp, corresponding to +10 to +11.6dBm at 50R.
4) At 40MHz the amplifier output is 2.0V and the 2nd filter 1.8V, suggesting that in fact the 2nd filter is beginning to bite.
5) At 45MHz the DDS filter output is just holding up, but the 2nd filter is cutting output very noticeably, probably mis-terminating the amplifier which now produces 1.7Vpp output, still +8.6dBm!
6) From 50MHz upwards the DDS LPF rapidy reduces amplitude at its output (and of couse the Amplifier input). The input and output of the amplifier have become negligible at 65MHz (15mVpp from the amp output is -33dBm) and zero by 70MHz, nothing visible at all on the 'scope. This corresponds well with the design characteristic of the filter! and is rewarding.

6m anyone?

The considerably improved output at the high frequency begs the question of whether the circuit would stand use at 6m, with an LO frequency of (52 + 10.7) = 63MHz. Although this was not within my original design aim, I believe that it would. I'm tempted to try it; the DDS output holds up well up to at least 70MHz and the amplifier has 'gain to spare'. Anyone wishing to try this would be advised to use a higher reference oscillator frequency than 150MHz, to push the first image frequency higher and make the filter less critical. The Butler is quite capable of 200MHz with a 'hairpin' inductor in place of the coil, but my own crystal was initially unwilling to work reliably at higher than its stated frequency (152 and a bit MHz). Robert, G3WKU, reports success with a 40MHz fundamental-mode crystal oscillating at 5th harmonic using a hair-pin 'coil'. M0RJD has now achieved 213MHz by selecting the 7th overtone of the 5th overtone crystal, by changing the damping resistor R304 to 1k. A separate oscillator could be used, as some are already doing. The AD9951 has 400MHz as the highest specified reference clock... 

If limited to a 150MHz reference, it is probably not worth trying to achieve 6m. The LO output needs to be 63MHz for 'high LO', so a filter corner frequency of 68MHz would be reasonable. The first image, which we must reject deeply, is at (150 - 63)MHz = 87MHz, so there is not much margin (we would aim for -60dB at say 80MHz or less, which is a bit tight and may not be fully effective when considering tolerances in the filter!). The required slope of the filter is also steep and may result in a poor response for in-band signals. The input impedance of the filter seems most badly affected with this filter architecture when trying to achieve a high slope, especially at 50R; don't neglect it!  A better solution for 6m is to use a considerably higher reference frequency. 175MHz should be achievable easily by the Butler circuit, if a suitable crystal can be found; it pushes the first image up to 112 MHz from 87MHz and gives much more filter design leeway, as well as reducing the sin(x)/x rolloff inherent in the DDS digital sampling. For higher Butler Oscillator frequencies, the resonant frequency of L301 in the tank circuit must be increased, so less turns (or even a 'hairpin') will be needed.

To put this into perspective, all these things are equally important regardless of how 6m is achieved. In fact, everyone so far that I know of who has extended STAR to 6m has also replaced the Amplifier with another design based on AD8008 in order to get adequate output - this may no longer be necessary, especially if the 2N3866 is replaced with a 2N5109 with higher fT.
I have no intention of stretching my STAR to 6m in the immediate future, so my filter design is only good for the standard STAR bands, but if anyone does this using a 50R filter I would love to hear from them.

Recommendation

My recommendation to anyone newly building STAR is to use the method I have described in order to greatly improve the LO output level. Others may not wish to rebuild the DDS filter on a working STAR; it doesn't make sense to do so (if it ain't broke don't fix it). The changes affect only component values; no extra parts are needed and every component already has a place on the existing layout. The increase in LO output easily exceeds the target +7dBm, even having takes steps to reduce it (a thing that not a lot of recent STAR builders need to do!).  I have shown why the earlier AD9850/9851 designs gave more LO output due to the DDS output architecture having no source matching, and explained why this architecture is not safe with the AD9951. The changes to the filter easily improve the LO output and are in no way a disadvantage to anyone using the AD9951. The new filter is neither needed nor recommended for use with the earlier DDS (AD9850) which has a maximum clock rate of 125MHz. AD9951 users could increase the clock to as much as 180MHz (5V supply), which would allow the new filter to be used, but it already provides higher output and the change is not required. New builders will automatically use AD9951, if only because of its better DAC (14-bit rather than 10-bit) and should seriously consider using the 50R filter in order to completely avoid LO amplitude problems.

Of course, if you are unlucky and do not get any benefit from this filter change, then although I will listen to what you say I can't do any more than that. Your mileage may vary. But do let me know what results you get!


The M0RJD filter at 50R
M0RJD's DDS, 50R Filter & Amp

Thanks to

I am grateful to  Glenn VK3PE, Duncan G4ELJ and many others who have spurred me towards this solution, and especially Robert G3WKU who has checked the calculations and generally joined in with the design and analysis process. Also profound thanks to Peter G3XJP, the 'father' of PIC-A-STAR. Without his generosity we would not have the STAR in the first place.

Mod details

Please remember that the above is Not an Official Modification, but do it just the same!

Now some of you will want to know the filter details.

The filter has a sharp cut-off starting at about 47MHz and reaching -60dB at 75MHz, which easily chops off the first image frequency of the DDS output. Above 75MHz, the attenuation worsens to 58dB at some frequencies.
For comparison, the original 200R filter cutoff begins at 43MHz and first reaches -60dB at 65MHz, worsening thereafter to  as little as -45dB at higher image frequencies.
The slower cutoff of the new 50R filter is necessary in order to reduce in-band effects, most notably a severe drop in input impedance. Since the first image with 150MHz clock is at 109.5MHz or above, this is very comfortable. Even the AD9950 with 125MHz clock has a lowest first image of 84.5MHz, still 10MHz in hand.
The greater stop-band attenuation of the new filter is another bonus, since higher images will be attenuated by at least an extra 10dB.

Component Values

These are the normalised, preferred values for capacitors and realistic targets for winding the inductors.
When winding inductors, I found that measurement showed a higher inductance than the calculated number of turns suggested. The strong recommendation is to measure your inductors and adjust the turns for the specified values; there are not many turns and it is easy to start with the calculated number and remove turns one at a time if necessary until close to the design values. I have given the turns I used on T-25-10 cores, but T-25-2 cores are probably good if the turns are adjusted appropriately. I wound my inductors with 26awg Tefzel insulated wire-wrapping wire, which is tough, easy to use, and fits, but almost anything that will fit will serve. Large diameter wire will ensure a high Q.

I have used the component references from Glenn's excellent ComboStar PCB design, since this is what I have built, and included the values used in the original 200R filter for cross-reference:

Parts List for 50R Filter and Amp
Component
Originally
New Value
Notes
C324
22pF
120pF

C325
33pF
150pF
160pF if available
C326
22pF
100pF

C3305
3.3pF
8.2pF

C3304
8.2pF
22pF

L3302
1uH
230nH
(9t 26awg wire-wrap wire on T25-10), (8t on T25-2 by SP5TAA)
L3301
680nH
190nH
(8t 26awg wire-wrap wire on T25-10), (7t on T25-2 by SP5TAA)
R3311
180R
33R

R312
56R
68R
I omitted this but it will aid correct filter termination & give some attenuation.
R316
5R6
10R

TT1
3+3:6t
3+3:3t
BN-43-2402, bifilar primary, ideally trifilar wound.

Final Circuit

This is an edited version of Glenn's ComboStar circuit, to show the new component values.
Final circuit

Reminder for new constructors

When winding toroids, remember the rule that 'each time the wire passes through the centre it is one turn'.
If the toroid is wound with the wire entering from one side of the centre hole and exiting from the other side, to make an inductor or transformer that stands vertically, this is very important and not intuitive; there is probably one more turn than it seems.
Toroid Turns: Not always obvious
At first glance, the toroids shown in the photo appear to be wrongly labelled, but the one on the left has two passes through the centre and that on the right has only one.

Follow-up

In the photo of my board above, you will have noticed that the Amp output transformer toroid L3303 has more turns than specified. This was as a result of early attempts at increasing output level, trying less and more turns. It made no difference, so I left it at 'more' turns so as not to remove & replace it excessively. In both photos, the flux residue from a million component changes can be seen, as yet uncleaned! Heatsinks will be fitted.  I have also fitted a 68R for R312 in order to pad the Amp input impedance. It reduces output amplitude from 2.0Vpp to 1.7Vpp at 40MHz, still an output of +8.6dBm. The measurements shown in the table are with no R312 present.

Reference Oscillator
Since the  measurements, my Butler Reference Oscillator is now running satisfactorily at the 7th overtone of the crystal rather than the 5th. This now provides approximately a 213MHz reference clock to the DDS, which runs slightly warm to the touch at 40MHz output. The inductor is a hairpin of wire-wrap wire in the photo, but is now a similar piece of heavier-gauge wire for mechanical stability.  The damping resistor R304 is 1k. The trimmer is set to a very slightly lower frequency than the crystal would suggest, in order that it starts up reliably.  This increased reference frequency would easily permit LO use at 6m, with a re-redesigned filter...

The 152.291667MHz Crystal is from 'anonalouise' on UK eBay and cost £5.00 with £2.00 delivery. It has a rather poor temperature drift, but once the oven has stabilised it is OK.  The crystal has 3 leads. The centre lead, which is welded to the can, I bent up along the can and it is not connected to ground. The can is in contact with the tab of the heater transistor TR304 which is decoupled to ground; it seems not to be troublesome. In the picture, the tab of TR304 is not cropped; this is because my original Xtal was in a larger can.

Butler oscillator running at 213MHz 7th overtone
The blue wires are the yet-uncropped leads of the thermistor.
The orange wire is the tank 'coil'.
152.291667MHz xtal gives 213MHz on 7th overtone

Send comments or criticisms

Is anyone actually doing this?

I will summarise the information that constructors provide, together with relevant comments. I have the authors' permissions to publish this. No permission, not published.

Robert G3WKU helped me through some calculations and kept my enthusiasm going! He has not used the 50R filter because his STAR is fully operational - but this is with his own modifications to the DDS Filter/buffer.
Robert says "I think it fair to say that you have well and truly 'nailed down' this problem once and for all."
Thank you for your considerable and welcomed help, Robert.

Witek SP5TAA has modified his filter as described, but using a trifilar-wound DDS output transformer and T25-2 cores (7t & 8t) with excellent results. He gets +9.2dBm at 29.7MHz with 10.695MHz IF. It would be easy to increase this level if necessary.
The cutoff is between 42MHz and 45MHz, with -34.5dBm at 65MHz.
Witek says "I think that your modifications are what all STARters should use in their builds."

Steve G7WAS is now using the 50R filter with trifilar DDS balun in order to overcome persistently low output and is now considering how to reduce output in order to avoid overloading the Buffer Amp! One idea is to use a splitter to provide some signal to a fron-panel connector to enable the DDS to be used as a signal generator. Steve is also investigating redesigning the filter for 75MHz. His PCB has suffered damage from a succession of 'amplitude improvement' mods. Previously, Steve found that changing from 2N3866 to 2N5109 actually reduced the output. He says "When I’ve had problems, with my Star build in the past, I found that it’s very easy to think it’s just me; everybody’s build works as designed. It becomes a bit disheartening, I find myself picking up little clues from various sources, that all’s not what it seems; others are having the same problem and have come up with all sorts of workarounds."

Dudley NO5L has measured the response of his build of the 50R filter using his N2PK VNA and says "It is flat with a 0.5db valley at about 25Mhz up to 42Mhz.  Then it falls to about -60db at 55Mhz and then rises up again towards 60Mhz; I can't check it above that.  But I'm sure it is as good as the prev. iteration of filtering, and I don't think I need to fiddle with things since the pass band is excellent below 42Mhz."