Broad Strokes

Broad Strokes

Hi all. Read on for part 2 of the same system we were looking at the other day. Interesting application as it’s a listening room, so we’re only concerned about the system response at a single point in space. This side-steps a lot of the spacial variance issues that are at the heart of sound system optimization for large venues, but there are still some basics we can take a look at. If you missed part 1, it is here.

I’m going to start by just looking at the left side on its own. There’s a great debate over whether a sound system’s frequency response should be flat or tilted, and then how much tilt, and what corner frequency, and on and on. For a variety of opinions, see this Roundtable article.

Also, check out this article on the subject by FOH engineer Jim Yakabuski. I am fortunate to call Jim a friend and we’ve had many conversations about this particular topic. For me personally it depends on what the system will be used for (speech or music) as well as the local acoustic environment (how much LF buildup the room will contribute). Ultimately this question will be answered by listening to reference recordings and tweaking the tilt until it sounds correct. Then that becomes the target curve, to which I’ll match the rest of the system. “Rest of the system” in this case means the other side.

The initial response actually had a slight upward tilt to it. Look at 3 kHz and up:

This is partly because the room is small enough that a lot of the HF bouncing off the walls and ceiling shows up in the measurement (notice poor coherence) and also partly because we’re only about 15 feet away from a relatively large loudspeaker that’s designed to throw a lot farther than across a living room, so there’s not nearly as much HF air attenuation as we would see further back in its intended application. This makes sense, but it sounded quite bad (as confirmed by listening to my reference music) so let’s fix that.

Pink = pre-EQ (raw response)

Blue = post-EQ.

Now, hold your horses. I’m sure you are wondering about that 63 Hz peak, and I’ll get there. I used a hi-shelf filter at 3 kHz to get rid of that upward tilt, and a cut at around 400 Hz to rein in that midrange bump. That might be a lobe from the crossover (it’s a 15″ system and the listening position is vertically below ONAX but 400 still seems low to me), or it might be a modal thing. The deep, narrow cancellations on either side lend credence to both hypotheses, and since they’re cancellations you can’t EQ them back up to normal (to a first approximation that is true). If it is modal, good acoustic treatment can do a world of good. But that’s not within the scope of the job.

Okay, so for the 63 Hz issue: this looks surprisingly similar to the “sub bump” that you see on a modern large format PA system out of the box. Since this system is primarily going to be used for the owner to play loud rock music, I left it. We’re running without subs here so this is actually an impressive showing. It didn’t sound resonant or honky during playback. If it ends up being too much we can always go back and roll it back, but I suspect it will work out well for the owner of the system. This is a judgement call, though, so there’s no right or wrong, just what’s best for the situation.

Usually, we also feed the output of the DSP into the analyzer so we can see the EQ curve on the screen as well. While this is extremely helpful, it’s not strictly necessary, and in this case it’s not possible since the DSP is onboard the active loudspeakers, so here we just see the before and after response. I’ll include a screenshot of the final filter settings from the DSP at the end so you can see the filters I used.

The applied EQ looks like a relatively minor change on the screen but boy, did it sound a whole lot better. I’m using 1/12 oct averaging, which is lower resolution than I usually prefer but because of the low coherence and nasty room problems, more averaging clarified the general trends for me in this case.

We’re dealing in broad strokes here, not trying to go in and equalize every single little bump. For a variety of reasons, the small-scale details tend not to be very audible, whereas the ear is incredibly responsive to the wideband stuff. For example, after listening, the HF still sounded too bright and one more dB on the shelf worked wonders. So by focusing on the broad strokes we can make some really big improvements with relatively few (gentle!) filters.

Note the phase response before and after is almost identical. The filters here are so low-Q that there’s virtually no change. This would not be the case for really narrow, high-Q filters, which are all too often adopted by the “autoEQ” plugins, etc. The phase shift itself is of course not audible, but the resonances certainly can be.

Now let’s go back and see the Post-EQ left compared to the right channel.

The matched phase traces indicate matched arrival times at the listening position (see part 1), and so looking next at the magnitude trace on top, we can see further issues: the right loudspeaker (brown trace) is significantly louder than the left side. If it’s within a dB it’s nothing to write home about but these graphs have a vertical scaling of 3 dB / division so what we’re seeing here is significant. (It’s also audible. It’s important to always check these things as we go. A simple listening test can prevent you from making bad decisions based on improper or incomplete data. It’s important to remember that the analyzer simply gives us more data, and it’s up to us to interpret that data and make an informed decision.)

So we need a gain change. How much? It’s easy to offset traces in the software until they line up – in this case it ended up being 4 dB. That’s larger than a manufacturing tolerance, so before cranking the gain in DSP it helps to actually determine what the issue is. In this case, one of the input gain blocks in the loudspeaker’s onboard DSP was set incorrectly, and after removing the attenuation, L and R lined up just fine.

Right’s just a smidge higher, but that’s okay because we’re looking at its unequalized response vs Left’s equalized response. Since most of the EQ work will be subtractive, this is nothing to be concerned about.

Right-side EQ choices were similarly sparing:

Red trace is post-EQ.

and here’s L and R together:

Considering the bad room, L and R are pretty well matched in a tonal sense. The EQ necessary for this was asymmetrical, because the room itself – as well as the loudspeakers’ placement within – is also asymmetrical.

Focus on the white line, because even bypassed filters contribute to the blue shading for some reason. Note that the macro trends are the same: gentle cuts in the lo-mid and the HF. I ended up with two cascaded HF shelf cuts to get it to sound right up top – again, the analyzer cannot provide this data. Trust your ears, but then look at the analyzer so we can correlate what we’re seeing with what sounds good.


Leave a Reply

Your email address will not be published. Required fields are marked *