H-Mode mixer

How I got involved

Around autumn 2005, I renewed my interest in HAM radio and especially home brewing receivers/transceivers after a very long period of other activities. I was quick to decide to design and build an HF all band transceiver. The plan was to use as much as possible proven designs freely available on the internet while maintaining a high degree of experimentation as that is what I like most of home brewing. I did not intend to build a kit.

The first version of the receivers front-end consisted of a preamplifier followed by an SL6440 Plessey mixer and a frequency stabilized (huff/puff) free running VFO. Soon the VFO was replaced by AD9951 DDS technology when I learned about that on the net. This was a real improvement with regard to controlling the receiver more in software but also because of the AD9951's excellent phase noise properties.

Although I had no means at that time to accurately quantify the dynamic range and intercept point properties of my construction efforts, I felt that there was a lot to be improved and that I needed a very different configuration without a preamplifier in front off the mixer. But with the SL6440 that was not possible because of the intrinsic NF of that circuit. It was not long before I stumbled on the CDG2000 homepage and started to study and appreciate the designs presented there including the H-Mode mixer circuit.

I build the H-Mode mixer, Manhattan style with trifilar FT37-43 transformers, moved the preamplifier behind the INRAD crystal filter and had a sensitive receiver most likely with an intercept point much better than with my earlier attempt! This was the point where the DDS spurs started to get my attention, as their level and numbers where really unacceptable. With all the gain inside and in front off the SL6440 they had been swamped into the noise floor, but now no more...

This was also the time when I first emailed Colin Horrabin, G3SBI, the inventor of the H-Mode mixer circuit, with a question about the input termination of the mixer in the CDG2000 front-end. I was puzzled by the fact that no effort was spent on terminating the RF-port of the mixer with 50 ohms consistently, while this was considered to be very important with the high-level ring-mixers to obtain the best IP3. Funny enough, although the H-Mode mixer is rather insensitive indeed regarding the intercept point and its input termination, the input termination issue came back to me in the light of the DDS spurs.

As I was going deeper into HF front-end design, I had bought a surplus spectrum analyzer with tracking generator in the need for a more solid quantification of the progression made or not made. While using the tracking generator as a signal source for measuring the MDS level of the receiver, I found that the spurs were much less present in numbers and level when I connected the generator through an 80dB step attenuator directly to the mixer RF-port instead of to the 4-pole band-pass filters normally in front of the mixer. The spur pollution was actually about acceptable in that configuration. A flat 50-ohm termination brought the spurs to their knees, although I had no explanation for this effect at that point.

To quantify this I recorded the available spurs on 15M as there were so many on that band. I recorded all reasonably loud spurs that I could find while tuning the DDS slowly with a step rate of 5Hz. That gave me 132 (!) spurs in total and the levels were measured in dB above MDS in SSB bandwidth with DL4YHF's excellent Spectrum Lab PC software. The average level of those spurs with the band-pass filter in front was 13,2dB (!) above MDS. Terminating the mixer's RF-port with 50 ohms reduced that level to -5,9dB. Inserting a diplexer between the filter and the mixer reduced the spur level to -4,7dB. At that point the solution looked simple: Add diplexers for each band, 9 or 10 total, and we are done.

Although the diplexers would have improved things greatly, I was looking for something better as the diplexers looked more like a workaround than to a solution to the problem. Furthermore they would add extra attenuation in front of the mixer and therefore directly worsen the NF. And there were some unanswered questions too. Why did many of the spurs reduce so much with the 50 ohm input termination, while still some were not influenced a bit and remained equally loud?

DDS spurs and the H-Mode mixer

I have come to the following categorization of the DDS spurs:

Direct spurs.
These spurs coincide with the IF frequency and leak directly from the LO-port to the IF-port. Obviously, there will not be so many of this kind and the huge amount of spurs that I have encountered could not be explained this way.
Direct heterodyning spurs.
These are strong spurs available at the LO-port that directly heterodyne with signals received at the RF-port of the mixer. These are the spurs that are related to strong input signals and disappear completely if the input signals are removed. Also this category did not explain all the spurs.
Self heterodyning spurs.
These spurs are caused by spurs that leak from the LO-port to the RF-port and then heterodyne with anything available on the LO-Port to produce signals at the IF-port at the IF frequency. This category produces vast amounts of spurs. All spur combinations in the full 200MHz AD9951 DDS output spectrum that are 9MHz apart (my IF frequency) will produce a spur within the IF bandwidth! This happens to be also the spur category that reacts to the mixer input termination. When a band-pass filter is connected to the mixer, all combinations in the stop-band of the filter are almost 100% reflected back into the mixer. With a 50 ohm resistive termination they are much more absorbed and their effect is much reduced.

What can be done about these different categories? Reducing the spur levels at the LO-port is the number one cure to category 1. This can be done in two ways: Band-pass filtering of the DDS output and increasing the clock frequency of the DDS to the limit. The lower the DDS output frequency with respect to the Nyquist frequency the better is the wideband spur free dynamic range (SFDR). As I operate the AD9951 at 500MHz not much improvement can be done there except waiting for the next generation of 1GHz DDS parts (AD9910 and friends). Better mixer symmetry reducing the LO-IF-port leak will help too.

The cure for category 2 spurs is the same as for category 1. Filtering of the DDS output. In addition, filtering of the mixer input will help too as less combinations at the IF frequency will be found if the input spectrum is limited as much as possible. Improving mixer symmetry will not help at all for this category.

The cure for category 3 spurs is: Filtering the DDS with a band-pass filter with a pass band smaller than the IF frequency. Any combinations that heterodyne to the IF frequency will then be eliminated. Furthermore, any improvements to the mixers symmetry will reduce the spur levels at the RF-port and subsequently the results at the IF-port. And last but not least, resistive 50 ohms termination at the RF-port will absorb the leaked spurs as well.

Because improving mixer symmetry really helps for 2 of the 3 categories and can be implemented at one place for all amateur bands, being the mixer, this is an approach that asks for it to be optimized as much as possible. The first successful steps in that direction were improving the transformers of the H-Mode mixer. I replaced my home-made FT37-43 transformers with Minicircuit TT4-1A used in the CDG2000 and observed lower spur levels. Next came Colin's tip to use the 74AC74E (DIL14) or 74AC74M (SOIC14) from Texas Instruments in the squarer as those parts have balanced propagation delays in contrast to the ordinary 74AC74. This again reduced the average spur levels.

H-Mode mixer and the bus switches

Around that time, Colin noticed that Fairchild semiconductors had not been sitting still after introducing the FST3125 bus-switch that made the H-Mode mixer so successful. Many new switches were available now, most noticeably the category of video-switches. A sort of drop-in part for the original FST3125 is the FSAV332 although in QSOP rather than SOIC package. FSAV332 looked especially promising as its datasheet states equal on/off times which could further help to improve symmetry. Also other switches looked interesting like the FSAV330 which contains a couple of SPDT type switches that could simplify the layout of the H-Mode mixer. At that point I wanted to have a more systematic approach and decided to test a couple of those switches in a fair way, such that the results could be compared.

Initially my comparison consisted of FSAV430, FSAV450, FSAV330, FSAV332 and the FST3125 as a reference. At that time I could only measure IP3 on 80M with 2 crystal oscillators with stock crystals about 100KHz apart. This worked quit well and I could measure IP3 well over +40dBm, but I had the nagging feeling that I was blind on what happened with the intercept point on other bands like 10M or even higher. So I ended up building two generators based on the AD9951 DDS to give me the ability to measure IP3 on any band and at any offset I wanted to. This was a little project on its own and it turned out not to be completely trivial to measure IP3 well above +40dBm on all bands. The higher the frequency, the more built-in IMD I encountered. The end-result is described in the next chapter of this document and is good enough to measure the H-Mode mixer well above +40dBm on all bands.

With the all-band IP3 measuring capabilities available a good comparison of the different mixers was beginning to get shape. It turned out that IP3 peaked much on 80M and 40M, where it is needed most, but would not maintain a > +40dBm level on the higher bands. So apart for trying to get good symmetry to fight the spurs a second goal became getting a mixer configuration that would do better than +40dBm on all HF bands.

FSAV332 turned out to improve over FST3125, but the SPDT mixers were disappointing. The one that had very good symmetry (FSAV430) had unacceptable low IP3. And the FSAV330 did not come close to the FSAV332. The hopes had been high for the SPDT type mixers also because the layout of the mixer was much more optimal than possible with FST3125 or FSAV332, at least on a single sided board with a ground plane.

All switches tested so far had in common that the mixer could be build with just one device. Not a surprise with the need for symmetry in mind. Colin again tipped off the NC7SZ384 and FSA3157 in the "analog-switches" category from Fairchild. They required respectively 4 and 2 devices to build a mixer with. Especially the FSA3157 was very good with spurs and IP3. Surprisingly this was an SPDT device! Because it was so good I decided to test this mixer with a fundamental frequency squarer, which is attractive as the SPDT switches do not need to be driven with complementary LO signals. This was also induced by the wish to keep the LO frequency as low as possible to get the best wideband SFDR from the DDS. The other reason was that one way or another, all SPDT mixers tested so far showed rather poor RF-IF-port isolation indicating that the mark/space ratio of the LO needed to be slightly different in this mixer configuration.

So I tested the FSA3157 with Colin's 74AC04 fundamental frequency squarer circuit which is a variation of the well known 74AC86 squarer, but without the complementary signal requirement. This turned out to be very good indeed. The IP3 remained at the good levels it already had, but the spurs could virtually be nulled out, although be it at the cost of poor RF-IF isolation. At this point the only goals not met were IP3 > +40dBm on all HF bands, and now also to improve RF-IF block.

H-Mode mixer and the transformers

Because quite a few different switches had been tested so far, the attention was going into the direction of the transformers now. Not in the last place because Harold E. Johnson, W4ZCB, recommended me to have a look at 1:1 transformers instead of the 1:4 transformers used traditionally in the H-Mode mixer circuit. Minicircuits have introduced a couple of less expensive wideband transformers looking promising given their insertion and return loss plots. The latest datasheets of ADTT1-1 and ADTT4-1 also showed good phase and amplitude balance data, which is interesting from a symmetry point of view.

Finally the ADTT1-1 transformer solved the IP3 and RF-IF isolation problem. IP3 is now well over +40dBm on all HF bands and still very good at 6M and even 4M! Spurs are at -9.3dB, the lowest level ever measured in this project, at the best possible RF-IF isolation around -60dB! Sensitivity to the bias point is also low. The price to be paid comes with a slightly increased conversion loss of around -0,4dB on most bands. Apparently at the 50 ohms level, the Ron of the switches is starting to play a role, but not much.

With the category 3 spurs so low now with this mixer configuration, it is almost a pity to have to implement a lot of DDS band-pass filtering to fight especially category 2 spurs.

After the overall success of the ADTT1-1 transformer, except for conversion loss, 4 more Minicircuits 1:1 transformers have been tested. These transformers are designed to be very efficient at very low frequency (>15..30KHz). Initially, expectations were low with regard to for instance IP3, but this turned out to be a surprise. These transformers were the best with regard to IP3, also on the higher bands. They all did >+40dBm on HF and some even on 6M! And last but not least conversion loss is very good indeed. The only downside is the rather modest spur reduction possible with this category of transformers. The winner in this category was the already long time existing T1-6T transformer.