H-Mode mixer

I almost gave up with the SPDT type mixers as they could not match the FST3125 or FSAV332 SPST mixers with regard to IMD. But when finally the FSA3157 was tested that did even better than the FSAV332 and FST3125, I was rather surprised.

Both the FSA3157 and the NC7SZ384 mixers did not suffer at all from the fact that the mixer is constructed from two to four separate devices rather than just one, with regard to IMD.

Initially it did not seem possible to build a mixer with the Fairchild switches that does better than +40dBm IIP on all HF bands. It turned out to be possible after all and that the transformers play a very big role in the IP3 picture together with the switches. The best values are found on the bands where it is needed most namely 80M and 40M. Those values are easily exceeding +46dBm making it very difficult to construct the surrounding subsystems (BPF and roofing filter) so that they don't compromise the mixer's IP3.

The 74AC04 fundamental squarer turned out to be a winning factor when it comes to getting rid of the spurs of a rather "spurry" unfiltered AD9951 LO-DDS. I am convinced that the spur reduction obtained with the non-adjustable 74AC74 divide by 2 squarer type mixers is influenced greatly by coincidental component tolerance/difference. This means that a good level of spur reduction, although possible, is probably not very reproducible with that configuration. The big advantage of the 74AC04 squarer is 2-fold. 1) It allows for a precise adjustment on each band. 2) It divides the LO upper frequency requirement by two, which is important with the AD9951. The wideband SFDR of this device is considerably lower below 40MHz than below 80MHz.

The FSA3157 in combination with the fundamental 74AC04 squarer and the ADTT1-1 transformer is very good when it comes to the reduction of spurs. Without any DDS filtering (except for its 200MHz anti-alias filter) the average level of 130 spurs on 15M was -9,3dB below MDS. This is pretty quiet! When adding a single simple 7-pole cauer low-pass filter cutting of above 70MHz, the average spur level drops to -10,3 dB below MDS. Actually a number of spurs got stronger with the 70MHz LPF, probably because this filter is inserted between the mixer and the ERA-1 DDS output amplifier. So the LO input of the mixer will not see 50 ohms all the time and possible some reflection will happen. The loudest spur to be found is +8dB above MDS and there are only 2 at that level. Now band-pass filtering of the DDS is only needed to defeat the so-called "heterodyning" kind of spurs as mixer symmetry does not affect those.

There is no need anymore to try to make the RF-port "see" a 50 ohms resistive termination at all times to reduce the spur level. No more half or full diplexers anymore in front of the mixer. This greatly simplifies things! And consequently no extra (although small) losses introduced by those diplexers worsening the NF of the front-end.

The nearly optimal PCB layout that is possible with the FSA3157 is also a big factor in the success of this configuration. The layout can be further minimized than the one used in this test, by placing the switches underneath the outer transformers. In that configuration all important switch/transformer connections are only millimeters apart. A 'production' version mixer could have 1 or 2 low cost SMD I2C 8 or maybe 10 bit DAC's on board to precisely set the bias point and the mark space ratio for best performance at the operating frequency. A single adjustment for instance for each 1MHz segment is probably more than sufficient and will not use much controller resources like CPU and ROM.

The category 6 mini-circuits transformers (ADTT1-6, ADT1-6T, TT1-6, T1-6T) turned out to be well worth investigating too. Although no winners in spur reduction, these transformers are the best if spurs are not the first concern. What these transformers have in common is that for an unexplained reason they almost completely eliminate the mixers bias point sensitivity. At the same time they also produce the best 3rd order law IP3 behavior. I never liked the steep IMD dips with especially the 4:1 transformers on 80M and 40M with respect to the bias point combined with the somewhat questionable IP3 3rd order linearity. These transformers solve that completely. Further more they exhibit low conversion loss on par with TT4-1A and very good, the best IP3 even on the high bands. These are the only configurations that do >+40dBm even on 6M! Their poor spur capabilities can be easily explained from some the spec sheets. At VHF these transformers show less amplitude and phase balance than ADTT1-1, causing asymmetry where the DDS produces the most spurs. Within the category of type 6 transformers the T1-6T is the very best. This introduces a difficult choice FSA3157/ADTT1-1/74AC04 or FSA3157/T1-6T/74AC04... In general the 1:1 transformers seem superior to 4:1 transformers in the H-Mode mixer application.

Giancarlo Moda's 2 transformer version is an interesting branch in H-Mode mixer development. Most impressive is the spur reduction possible with home made transformers with this configuration in combination with the FSA3157 switches. Furthermore experiments with the 'mismatched' version of this mixer raises questions why the IMD behavior is so much improved by using a 1:1 output and 1:4 input transformer. For some the extra insertion loss around 2dB, introduced by this mismatch, will be unacceptable.


Without special care to the input band pass filters and the roofing filter that surround the mixer, the overall IP3 of the complete frontend can be much lower than what can be reached with the H-Mode mixer alone. In other words: it takes a lot of determination to make the mixer again the weakest chain in the receiver. Both quartz and iron-powder can cause considerable IMD, when aiming for say above +40dBm, often with bad non 3rd order law behavior.

The solution with the roofing filter is found in high quality crystals that have been manufactured in such a way that surface pollution is minimal or absent. Cheap mass-produced "Asia" digital oscillator grade quartz is not produced to such standards and simply does not have the quality to be used in a high-end roofing filter.

With respect to the input band pass filters, the key in minimizing the IMD is found in minimizing the magnetic flux inside the core. Any desire for really small and compact BPF's is not very productive as at least on 40M only filters made with the big toroids show minimal IMD. Also wider filters produce less IMD internally, causing an obvious trade off.

I find it quite amazing that on 40M toroids with a size usually used in 100W PEP transmitter output LPF's are required to remove the IP3 bottleneck in the receiver.

It turns out to be quite possible to construct a frontend with a dynamic range in excess of 120dB in SSB bandwidth on 40M. The best news maybe is that it is possible with components that are readily available today. No impossible to get specialized surplus parts are needed. Everything is still off-the-shelf, also the special QT crystals used in the roofing filter. However good quality polystyrene capacitors may become more and more a problem as many manufacturers have stopped production of those capacitors as far as I know years ago. Most certainly alternative good RF grade caps will remain available, although probably a bit expensive.

The biggest improvement made in the frontend regarding CW reception is found in a 500Hz wide 9MHz roofing filter. It turned out to be possible to use the so successful quadrature hybrid connected filters concept also for a 500Hz wide filter with low losses using the QT crystals. This gives the 9MHz down-conversion frontend quite an edge above the now dominating up-conversion frontend approach in most commercial high-end rigs. At low VHF (40MHz - 90MHz) where up-conversion 1st IF's usually are located, this kind of roofing filter performance seems not possible at present time!

Back to the TOC