[arghx7]

I don't know exactly what's going on here, but I can at least post about a few things that I've had some experience with. I'm not an NVH expert but I've had some exposure to it on a system design level and on an experimental/testing/optimization level. So I can suggest, for academic purposes at least, an experimental way to approach the issue.

So we know that exhaust makes noise, motor mounts and loose objects make noise, that kind of thing. The main things that are specific to a modern n/a DI engine are:

1) sound of the direct injection solenoid valves
2) sound of the high pressure fuel pump and fuel pulsation
3) sound of the hydraulic lash adjusters and valvetrain

There are a lot of things that affect engine NVH on a modern direct injected engine, but you can't figure out a root cause on a transient WOT from a chassis dyno like that. What you really need to do is put the engine into an NVH lab engine dyno. First thing I would do is motor the engine in 200rpm increments. Now what the hell does that mean?

Does this noise occur when the engine is actually firing, from a combustion event, or can you hear it when it's just spinning (like on a decel in the vehicle)? An electric motoring dyno bolts to the flywheel and spins the engine up and down to hold a speed. The engine doesn't have to run, but you can communicate with a prototype ECU in realtime and start commanding different AVCS positions. The beauty of that is that you can isolate whether the noise is coming from some valvetrain issue or fuel system issue without any combustion occurring. This is important because the pressure rise characteristics have a huge effect on engine NVH, especially on DI gas engines and diesels. Faster pressure rise=more noise, and that's one of the reasons why diesels are so loud.

So first I would motor the engine and see if I can audibly hear the sound, and see what my NVH instrumentation can pick up at the same time. Figure out the frequency and order of the noise in a motoring condition. Perform cam phasing sweeps at each steady-state rpm point and record data with different levels of intake and exhaust phasing. Generate plots of cam phasing vs various noise parameters (decibals, frequency, etc). Then repeat the test with the engine firing--actually run the engine at full load. See what the influence of combustion would be, and also perform cam phasing sweeps and fuel pressure sweeps to see what the influence of the high pressure fuel pump is.

Finally, I would feed that information to a simulation specialist who can do finite element analysis and figure out what the heck is making the noise. NVH simulations make little nifty animations that show the stress and strain on various engine components under different conditions, and can assess potential NVH. Then from all that work devise a countermeasure with hardware (motor mount, special engine cover, etc).

If a hardware countermeasure is not possible, look at your cam phasing sweeps, look at your NVH data, and determine what is an unacceptable level of NVH. Figure out which set of engine parameters fall create the unacceptable level of NVH. Weigh that versus engine performance and combustion data, and make a judgment to create a calibration-based NVH countermeasure in lieu of a hardware change.