2002 Subaru Outback 3.0 Rough Idle

I have a 2002 Subaru Outback 3.0 (6-cyl) with a rough idle.

It evens out at higher RPM.

No codes, but sometimes will read a P0420. Rear O2 performs as expected with WOT test and front AF readings.

Tested both front wideband AF sensors under WOT - behave as expected.

Only suspect PID is a higher LTFT on Bank 1. 10% at times for a total trim of 15% when added to STFT. Bank 2 is much closer to zero.
LTFT decreases with throttle 

Done so far:
Fuel pressure test per Factory Service manual - passed.
New Spark plugs NGK brand (were due anyway)
Scope of secondary ignition with COP paddle probe and picoscope (Looked fine but very hard to get a good waveform with the COP probe)
Checked for large vacuum leaks with spray bottle
Replaced a few deteriorating vacuum hoses but more remain.

Where should I head next? I am really thinking it's a vacuum leak but this is a speed density engine (MAP only) and I know they tolerate small ones pretty well.

Injector waveform on Bank 1? 
Vacuum leak test with smoke machine? (would need to purchase one)
Relative compression test?
Injector flow test?
 

I would probably go after compression or relative compression testing next to verify the engine is sound. Idle only misfires can be valve sealing related.

I finally got around to performing the relative compression test last night. I pulled the fuel pump relay on a cold engine as the Subaru does not disable injectors with WOT. 

I used my PicoScope connected to a Hantek high current clamp.

My compression peaks and troughs are all similar but there are some odd double humps on the compression rise. Any idea what these are? 

 

Sorry for the late reply, looks like just the amp clamp picking up some noise in my opinion.

My compression peaks and troughs are all similar but there are some odd double humps on the compression rise. Any idea what these are?

It is not uncommon, in my experience, to see a Relative Compression waveform that have a "plateau" at the top of the compression stroke. I have never seen or heard anyone explain what it is but, I have my own theory. I may, or may not, be correct.

A relative compression test is, basically, graphing the "work" exerted by the starter. As the crank shaft turns, the connecting rod that is attached to it is on one side of the crankshaft as the piston ascends on the compression stroke. Then, for a very brief moment, at TDC, the piston takes a pause as the connecting rod makes the transfer to the other side of the crankshaft, and the piston begins its descent on the power/expansion stroke. I believe that the "plateau" (marked in GREEN) is because, at that moment the starter gets very brief moment of rest before it has to "drag" the piston back down. Is the piston being "dragged" down by the starter the reason for the second side of the "plateau"? I don't know but, that is my theory.

Your waveform, however, has an EXTRA hump (marked in RED) that I am not use to seeing. It seems to be every other cylinder, suggesting a bank-to-bank issue.

 

Going back to the principle of graphing the work that the starter is exerting, what would cause the starter to get a "little break" DURING the compression stroke?  My first thought is compression loss, which brings me to adding a filter to a Relative Compression waveform.

Many people are opposed to adding a Low Pass Filter to waveforms because they are afraid that "you will filter out detail". As Noah pointed out, the waveform contains a lot of noise. In my opinion, that noise is HIDING the detail. Look at the raw waveform, with all of its "detail", then filter it until it yields some information of value.

First, understand what the filter does.  It filters out activity that happens faster than the set frequency. The noise in that waveform is happening at a frequency of many thousands of times per second. Relatively speaking, a cranking frequency of about 200 RPM, or 3.3 rotations per second is VERY slow.  You do not need to sample thousands of times per second.

I, usually, filter the hell out of my Relative Compression waveforms. I will set a filter of about 100 Hz, or slower. Anything that is happening faster than 100 times per second, is just noise. Get rid of all that noise, so that you can compare the actual WORK of the starter, rather than the noise. You just might see a subtle bank-to-bank compression difference.
 

Good eye Chad. I will admit, I did not notice that repeating dip on every other hump. Curious to see how it plays out