Warren WA8TOD previously raised concerns about the level of intermodulation being generated in the µBITx TX signal path.  IMD may not be as pressing an issue as harmonics and spurs, but the µBITx seems to generate more IMD than it should. This can cause audio distortion and splatter (including out of band transmission if you are operating near a band edge).  IMD only occurs in SSB modes, not on CW. If you are CW operator you only need to address spurs and harmonics for which there are now solutions.
General expectation about intermodulation
The generally accepted limit for SSB intermodulation products is a minimum of 24 dB between the lowest of the twin tones and the highest of either the third order (products immediately adjacent to the twin tones) or fifth order (products next over above and below the third order) products.
Warren’s observed IMD
Warren’s µBITx shows -12 dB at 3.6 MHz. Conditions are: 30 mVrms audio input and RV1 set for 5 watts RF output through a 4 MHz LPF. His board has the onboard PA filters removed and strapped and has the additional 45 MHz filter with 12:1 output impedance transformer in place of R27.
Where is the IMD being introduced?
In order to better understand where the IMD is being introduced in the radio Warren started all the way back at the balanced modulator output and measured IMD at thoughtfully provided test points (on the v4 main board) up through the driver output. In general, once IMD products are introduced at early stages they tend to only get worse as the signal progresses down the chain. The key to fixing it is finding the root cause as early in the process as possible.
TP17 is the output of the balanced modulator and the 12 MHz SSB filter. IMD products here were below the noise floor of the measurement configuration.
TP16 is the output of the first bi-directional amplifier and IMD at this point measured -35 dB, already much too high and and indication of non-linearity in the amplifier that must be addressed.
TP14 is the output of the onboard 45 MHz filter following the 2nd mixer. This actually shows a slight improvement but probably within measurement error at -37 dB. This measurement pretty much exonerates the 2nd mixer as a significant contributor to the IMD issue.
TP16 is the output of the second bi-directional amp and again there is a serious deterioration in IMD with the amp adding 11 dB to the problem.
TP1 is the output of the 2nd transmit mixer (labelled 1st mixer in the text) and is the first time we see a signal at air frequency of 3.6 MHz. The mixer added 5 dB of IMD to the total… too much and probably indicative of low injection levels as has been stated in the past. On the other hand, it is not the primary culprit by far.
TP3 is the output of the first pre-driver and of RV1 and it adds a little over 1 dB of IMD. The total IMD at this point is 5 dB less than the acceptable amount and it is only beyond this point that we are able to control power levels with RV1 which would normally be the adjustment point for controlling PA IMD. In other words we are starting out with an unacceptable signal from the low level stages and only now getting to where IMD is normally introduced. A contemporary radio would show normally show IMD levels at -45 dB or better at this point.
From this point forward, at the five watt level, the combination of pre driver, driver, and PA added 5 dB of IMD. This amount would be perfectly acceptable in most radios starting out with clean drive and would allow the total power to be increased by RV1 adjustment to significantly higher levels.
Cause of the IMD
The IMD problem is rooted first in non-linearities in both bi-directional (bi-di) amps and then in both mixers. The focus is likely to be initially on addressing concerns about linearity in the bi-di amps (on TX only of course).
Possible improvements to the bi-di amps
Glenn VK3PE notes that the original article by Wes Hayward and Bob Kopski on bidirectional amplifiers shows a slightly different biasing arrangement and feedback in the first stage compared to that used in uBITX. As shown, gain is 15dB and flat to 100MHz within 1dB.
A more conventional resistive voltage divider is used and the feedback is AC only. Two extra parts are used. The article doesn’t mention IMD though specifically, as a performance target or measure it.
Glenn has plotted the gain difference between uBITX and Hayward versions of the bi-di amplifier after building a prototype.
There is a difference in input levels before output clipping occurs. The biasing arrangement also gives different Iq. Gain is reduced about 4dB in the Hayward version over the uBITX. It accords closely with test results from Hayward’s paper of 15.5dB. Glenn got 16dB of gain at 30MHz.
Haywards paper gives some values for varying the gain to other values also, so there is more experimentation to be done.
Henning DK5LV notes that the designer of the Bi-Di amp states his version is designed for 15 dB gain, which is why he has a series feedback system (680 ohms + cap) from collector to base, and the biasing is done with two extra resistors.
Ashar VU2ESE in designing the µBITx uses only the biasing and feedback with the two resistors.
Henning suggests that the result is that Ashar’s amp has about 20 dB gain, but the frequency response is worse due to the limited fT of the 2N3904 and input reflection (S11).  Due to the higher gain, IMD must be worse for the same input level. If the textbook curve applies (3 dB for every 1 dB of input power change) the IMD will be 15 dB worse.
Raj VU2ZAP has found good reduction of spurs with the following mods:
In considering adding a new LPF module to replace teh original on board LPF matrix, it is helpful to remove the audio switching from K3.  This reduces RF signal routing on the uBITX board. Many have found the signal routing via K3 causes RF to get into the audio so this represents good practice.