This seems to be the quarter for speaker matching.
Last week, I trundled off to Nova-K studios with an audio analyzer and a truck full of surplus acoustical treatments. My mission was to figure out how much improvement in the monitoring environment could be made ‘on the cheap’.
I had already recorded several projects at this particular studio. I knew that it was a challenge to create mixes there that translated well to other systems. After living with the situation, we finally reached our breaking point one day.
The smoking gun
One one tune, we wanted to add some room sound to glue the electronic sampled drums together as a cohesive entity. In a stroke of madness, we decided to reamp the drum tracks through the control room monitors, and record the resultant sound using a matched pair of small diaphragm condenser mics. Upon playback, the sound was unusable due to a nasty resonance. In fact, the resonance was so pronounced that a guitar tuner program built into my phone had absolutely no problem in identifying the resonance as 67.some-odd Hz.
We tried reorienting the mics & monitors, trying different mics, etc. We could not eliminate the resonance. Further, now aware of its presence, we could clearly hear its effect in the mere playback of almost any program material.
A quick back-of-the-envelope analysis of wavelengths, room dimensions, and modal theory revealed that this resonance was unsurprising. We will return to this theme in the future. For the purposes of today’s post, we need to move forward a bit.
Matched pair? Riiight.
So I arrived at Nova-K with my analysis rig, and set it up. I figured I’d get some raw shoots of the current situation to establish a baseline before tweaking anything.
So I set my measurement mic up in the mix position, and proceeded to measure the output of the left channel. There were significant deviations from flat response, including the expected huge peak in the 68 Hz region. We measured a similarly dismal response in the right channel. Further, the left side was weaker by 4 to 8 dB through the range from about 800 Hz – 8 kHz. This seemed extreme, but plausible given the lath-and-plaster nature of the room, and its asymmetries.
We moved the mic in turn to approximately 1 foot in front of each monitor in turn, expecting the left-to-right deviation to all but disappear. We were surprised, but not completely astonished to learn that the discrepancy from side to side in the upper mids to remain about the same.
Starting to question the speakers (as opposed to the room), we swapped the left speaker to the right mount and the right speaker to the left mount. Measuring each again showed that the upper midrange difference moved with the speaker, rather than staying with the location in the room.
Now pretty convinced that the difference was in the speakers, we wanted to prove this. We placed them at chair height, right next to each other, and baffled them off front, sides, and top from the room with heavy gobos, and in the rear with blankets. Placing the mic about one foot in front of the seam between them, we measured again. The following graph clearly shows the discrepancy in response — the ‘left’ speaker is in orange, and the ‘right’ speaker is in blue.
One can readily see that, save for a strong peak at about 1300 Hz, the left speaker has a significantly weaker response throughout the upper mids as compared to the right speaker.
The Speaker Scrutinizer
It was evident that, given the speaker to speaker inconsistency, this was a problem that needed addressing before the room treatments. Accordingly, we changed tack.
The speakers were model PC2 from Phase Technology. Other then hearing them at Nova-K, I had no knowledge of this manufacturer. While not marketed as ‘studio monitors’, per se, further investigation revealed that they have a decent reputation in audiophile communities. These are a two-way, bass reflex design, with ~ 6.5″ kevlar woofer with a flat plate, and a 1″ soft dome tweeter. These had been in service at Nova-K for approximately 13 years.
So we started disassembling one. This disassembly revealed a rather solidly built unit, with a hefty woofer, and a surprisingly complex crossover network for a two-way bookshelf system:
A visual inspection tuned up nothing amiss. So we tore into the other unit. We were surprised to find that the rather substantial internal bracing from the first unit (s/n 01290A, previously ‘left’) was missing from the second unit (s/n 01021B, previously ‘right’), and that the second unit’s XO was mounted at a rakish angle.
Again, a visual inspection turned up no issues with the right speaker that would seem to explain the discrepancy in frequency response. It was obvious that we needed to measure each individual component to identify the source of the discrepancy. Lacking a full array of measurement gear, I took the speakers back to Rocket Surgery Labs (a subsidiary of my sound company q music inc.) for a more detailed analysis.
On the bench
Back at the Lab, it was time to characterize each component, looking for differences from unit to unit. The speakers each consist of essentially three components — the woofer, the tweeter, and the crossover network (XO). Access to each of these is through the woofer’s mounting hole.
Reasoning that there were no ‘strained’ sounds from any of the drivers indicating damage, and that the XO was rather complex for a two-way design, I started my investigation with the XOs. I pulled them from the cabinets. I decided to characterize them using a process similar to that described in this previous post. This process employs SMAART, as a dual channel FFT, to generate a Transfer Function for each XO output. See the aforementioned post for more info on this process.
Here is an overview of the test setup:
Before setting up the test rig, I measured each of the woofers and tweeters at a DC resistance of 2.7-3.2 ohms. Using faulty reasoning that I now discount, I assumed these to be three ohm nominal drivers. In retrospect, these are likely four ohm nominal drivers. However, as we will see, this is immaterial for the purpose at hand.
Accordingly, I loaded down the LF output of the XO with a 3 ohm wirewound load resistor, and the HF out with a 3 ohm load composed of two 2 ohm resistors in parallel, for a resistance of 1 ohm, in series with another 2 ohm resistor, for a total of 3 ohms. Here is a closer look:
In this photo, one can readily see the output of the amp on the gray-sheathed red and black wires, connected to the XO inputs (white and black), the XO LF outs (blue and black) connected to the green wirewound resistor with the alligator clip leads (red and black {offscreen}) and the XO HF out (red and black) connected to the resistor network through the alligator test leads (yellow and green at top). Also attached to the LF outs is the Interface’s DUT input, through the red and green alligator leads in the foreground.
In the process of this testing, I discovered a cold solder joint on one of the inductors. After repairing this, the XO’s Transfer Functions were remarkably similar:
Remarkably similar Transfer Functions. I next reassembled the units, but included leads from each XO output <> driver connection, each passed out the bass reflex port. This allowed me to measure electrically at the crossover outputs with the complex load of the drivers and the cabinets, rather than the simple dummy load. Here is how this looks physically:
While the photo is blurry, you can see the tape flags labeled LF on the red and black pair, and HF on the yellow and green pair.
With the real load of the speakers on the XO’s, they still appear well-matched:
Both amplitude and phase are very well matched, for both LF and HF, from unit to unit. Accordingly, we can rule out the XOs as the source of the frequency response aberration.
For the interest of completeness, the following graph compares the XO TFs in both the case of dummy load, and of actual drivers (and cabinet acoustics) as loads:
We can see that the real load of the speaker drivers, as well as the reflected acoustic impedance of the cabinets, have a non-negligible effect on the crossover transfer functions.
Woofers
I next decided to move the L woofer to the R cabinet and vice versa. If the aberration went with the woofer, then the woofer would be implicated as the differing component. If the difference stayed with the rest of the unit (cabinet, XO and tweeter), then the woofer would be absolved. This would require acoustic measurements. Further, I would need to re-baseline the measurements due to the different acoustical environment in which I would be testing.
I placed the speakers side-by-side on the corner of the workbench, and carefully positioned my measurement mic to be equidistant from the axis of each driver. Here is an overview:
You can see the similar configuration of the SMAART test computer, the Interface, and the Power Amp. The DUT input of the interface, however, is now being fed by the measurement mic visible in the foreground. The next picture shows the view down the axis of the mic:
One can see the test leads from a previously described test exiting the port one of the speakers. These are not connected in this test.
With this rig, I made baseline measurements. These are shown below. The left speaker is in orange, and the right speaker is in magenta:
Hey! In this acoustic environment, the right speaker is impressively flat – within +/- 3dB from about 50 Hz to about 18 kHz. The phase response is also very smooth.
One can clearly see that there is still a significant loss of amplitude in the upper mids in the left speaker. Interestingly, the 1.3 kHz spike is no longer present. Was this a byproduct of the cold solder joint on the XO? More on this later.
I then swapped the woofers from cabinet to cabinet and remeasured. The problem did not move with the woofer — it stayed with the rest of the speaker. Here is the curve for the right speaker with the woofer from the left in green, along with the right speaker with right woofer from the above measurement in magenta:
And similarly, replacing left’s woofer with that of the right yields similarly poor performance for the combination. Here is the earlier left measurement in orange, with the new measurement of left with right’s woofer in blue:
As the problem stayed with the system, and did not move with the woofer, we conclude that the woofers are substantially identical. I returned the woofers back to their original systems.
Tweeters
As the XOs are identical, and the woofers are identical, the difference must be either in the tweeters or the cabinets themselves. I prepared to swap the tweeters from cabinet to cabinet to make another measurement. Upon removing the diffraction foam on the face of the baffle, I was met with quite a surprise. The tweeters are of substantially different design! See the photo below:
The tweeter on the left appears to have a trim ring around it, pressed into a recess in the face of the baffle. The right tweeter has an integral mounting flange. Investigating further, I snapped some photos from the inside of the cabinet at the backs of the tweeters. Here is the left:
Note that the tweeter itself does not extend through the baffle. It is merely screwed to the front of the baffle, and the speaker wires are passed through, and sealed with modeling clay.
On the other hand, the right tweeter does extend through the baffle:
This tweeter appears to have a back chamber, adding perhaps an inch and a half of depth behind the face of the tweeter. No wonder it sounds different.
C’est la vie
Despite being sold as a matched pair (!), they are of considerably differing design. Oddly, the presumably newer one (based upon serial number), with the additional cabinet bracing, has the poorer tweeter configuration. Ideally, we would procure one of the old style tweeters for the other speaker, thereby bringing it up to performance with the better one. However, these are thirteen year old speakers. Accordingly, sourcing may be a problem. While attempting to locate a suitable replacement, we still have mixes to do. Accordingly, we will accept the differing performance for now, and patch around it as best we can with eq. As the phase looks rather smooth through this region, we expect to be able to largely mask this difference in performance with some tuning. At least we eliminated the nasty peak at 1.3 kHz.
The room tuning will be the subject of my next post…