Rockin' with Joe Bear

The EAW KF650isfx

So a while back, my sound reinforcement company Rocket Surgery Sound (a division of q music inc.) bought five used EAW KF650isfxP speakers. These are true professional-grade three-way arrayable speakers for live sound reinforcement. The KF650- is the base model name, the -P suffix indicates they were made for permanent installation, the -is- indicates the first iteration of these speakers, and the -fx- indicates that they were a special Build To Order version with an integral three way passive crossover, meant to be operated full-range only.

The standard product most like my speakers are the KF650is (without the -fx-). These standard units are meant to be either bi-amped or tri-amped. So rather than the single crossover (XO) of the standard product, the KF650isfx has three XO networks spread across four separate modules. The woofer is filtered by the LF XO module, the tweeter is filtered by the HF XO module, and the mid is filtered by the MF network, which is spread across the MF XO module and the IO panel.

Problem and Potential Solution

Upon receiving these speakers, each sounded radically different than the other. Cabinetry seems identical. The drivers all seem roughly equivalent — though I will characterize them fully in a future installment. Apparently, the difference in sound from speaker to speaker is due to variation in the XO networks.

In order to restore the proper operation of the XOs, the plan is to characterize each network’s Transfer Function. With the assumption that at least some XO networks are in good shape, I should be able to determine which XO networks have drifted from their factory values, and then drill down to component level troubleshooting to identify the failed components.

I have access to a tool called SMAART, which is a computer program that measures and compares audio signals. Using SMAART, I can compare the relative amplitude and phase, between the input to an XO and its output, across the entire audio frequency range. This generates a Transfer Function.

Accordingly, I will capture a Transfer Function of each XO in turn, and compare these for any anomalies.

Test Setup

The following is a diagram of my test setup:

XO test setup diagram

Note that I am using an 8 ohm resistor to load the output of the XO. Each driver in the KF650is* is an 8 ohm driver. While a speaker driver is a more complex load than a resistor, the resistor will serve for this characterization. Recall we are just needing to compare Transfer Functions from XO to XO to look for differences.

The rather low impedance requires us to employ a power amp to drive the XO. The power amp is being driven by an audio interface, which is fed a signal from SMAART.

The inputs to the interface, and thereby the inputs to SMAART, are at the input of the XO (V(omega)in) as the reference signal, and at the output of the XO (V(omega)out) as the measured signal.

SMAART is able to perform an FFT (Fast Fourier Transform) on both the reference signal and the measured signal and compare them. This generates the system Transfer Function

H(omega) = V(omega)out / V(omega)in

The following is a photo of the test setup:

XO Test Setup

Here, at the left you can see he computer running SMAART atop the amplifier, the interface in the center, and one of the XO modules on the right. The load resistor is the green cylinder behind the XO. There is also a DMM in the picture which may come in handy. Last, there is an assembly drawing and a schematic for the XO.

Low Frequency XO

For no particular reason, I started with the low frequency (LF) XOs. I found that four out of the five LF XO networks matched very closely. See this following for a Transfer Function (TF) graph:

LF TF - 3203, 3204, 3206, 3207

This graph shows all four TFs match within a fraction of a dB for all frequencies of interest.

The following is a higher resolution view of the same:

LF TF - 3203, 3204, 3206, 3207 - hi res

I was quite pleased that four of the five matched so closely. Indeed, even the phase trace is as tight as I could hope for. Phase only starts to deviate around 1 kHz, but the signal is already in excess of -50 dB down by this point, essentially being inaudible.

With such tight correlation, I am quite sure that these XOs have not drifted appreciably since leaving the factory. This is testament to the quality of truly pro level gear.

It is worth noting that the graphs show approximately -1.5 dB of insertion loss for the LF XO. This represents a waste of power. Converting to bi- or tri-amp operation at some future date could regain this lost amplifier power.

But not all is perfect

Unfortunately, the fifth LF XO (s/n 3205) did not fare as well. The following shows this TF superimposed upon the other four:

LF TF all

Note the discrepancy of the orange trace. This looks as if the main XO network 1s operating properly, but a second tuned filter at about 250 has failed.

By performing some resistance and capacitance measurements of the failing XO, compared with that of the good ones, I determined that L10 had failed on the LF XO 3205. This is a 0.9 mH inductor. The failed unit displayed a DC resistance of 0.2 ohms, while the healthy ones had a DC resistance of 0.6 ohms. Visual inspection suggested that the failing L10 may have overheated, melting insulation, and shorting the coil.

To verify that this was the only failed component I swapped L10 between the bad XO 3205 and the good XO 3203. The previously good 3203 with the bad L10 now failed:

LF TF - 3203 with bad L10

The orange trace is the first measurement of the failed XO, while the blue is the previously good 3202 with L10 from the failing unit.

I next captured a TF for the previously failing XO 3203, with L10 replaced with one from a good XO:

LF TF - previously bad XO with good L10

Note how closely the ‘repaired’ XO (orange) matches one of the four matching traces from above (blue).

Conclusion – a new L10 will restore LF XO 3203 to proper operation. At that point, all five LF XOs will be effectively blueprinted. I sure hope EAW can provide these inductors.

Mid Frequency XO

The MF XOs are somewhat more involved, as they are spread across two assemblies – the IO panel and the MF XO proper.

I first used the IO Panel sn 3203 as a control, and measured the five MF XOs.

As the response of the MF XO does not extend to DC, SMAART must compensate for the concomitant delay. I started by taking an impulse response:

MF Impulse response

And set SMAART’s internal delay to the corresponding .41 ms.

I then measured all five MF XO’s, using the same IO panel. Much to my delight, all five MF XOs exhibited tight correlation:

MF TF - all five modules

There is a little phase difference creeping in around 100 Hz on the low end, but the amplitude is already minimal by that point.

IO Panel woes

Unfortunately, the picture was not so rosy for the IO panels. Using the MF XO 3203 as a control, I measured each IO panel in turn. As I will explain later, I determined that IO Panel 3203 was one of the ones in good shape. As I used this IO panel for the MF XO measurements, this meant that I did not have invalidate all the MF XO TFs.

The following shows all five IO panel TFs superimposed. from the picture, one can see that there is a group of two, another group of two, and a third (fifth) outlier.

TF all IO panels

There were two IO panels that exhibited proper operation — 3203 and 3206. Their TFs are displayed below:

IO Panels 3203 and 3206 -- good

Excellent correlation.

The other group of two are as follows:

IO Panels 3204 and 3207 -- bad

Visual inspection revealed that a capacitor lead had been clipped on these two IO panels, essentially removing them from the circuit. Jumpering these leads caused the unit to whine loudly. Using a DMM and capacitance meter, I determined that these clipped capacitors were faulty. Some previous tech’s means of ‘fixing’ the units after failure was to remove them from the circuit. This removal is the cause of the radically different TF.

These caps are 15 uF / 100 V bipolar electrolytic. As two out of the five have failed, it seems likely that this is a weak area of the KF650isfx design. It would seem prudent to increase the voltage rating of this cap to 200 V or so.

The last IO Panel is worse yet:

IO Panel 3205 - dead

Note that, at the middle of its passband, the TF is essentially dead – the level is attenuated in excess of -50 dB.

By measurement, it was discovered that the other cap was faulty. This is a 30 uF / 100 V unit. Again,would seem prudent to increase the voltage rating of this cap to 200 V or so.

High Frequency XO

I performed an Impulse response on the HF XO:

HF Impulse

After setting the delay to 0.02 ms, I captured TFs for all five of the HF XOs. Much to my delight, all five exhibited excellent correlation:

HF TFs - all

In the process, I learned that the HF output wires are unconventionally colored. The black wire is hot and the red wire is ground. By hooking the colored wires to the correspondingly colored terminals on the tweeter driver, one obtains a 180 degree polarity flip. This is likely an intentional design artifact.

Unfortunately, before I realized this, I blew up my amp by connecting its output hot to the interface’s ground. Oh well.

Summary

Well, if my analysis is correct, it should be easy and cheap to restore the crossovers to factory fresh operation. Three capacitors (cheap and easy) and one inductor (moderate expense, and perhaps difficult to find) are all that is required.

I will follow up some day with a post-repair report.