Denon AVR-3806 Audyssey MultEQ XT Test Report
Audyssey’s MultEQ XT is a sound equalization system that’s finding its way into more and more audio-related products destined for the consumer electronics marketplace. Its encapsulated on a TI Aureus chip, found in Denon flagship products or in the case of the Denon AVR-3806 and other brand products , an Analog Devices’ chip. Based on research originally carried out under a National Science Foundation grant at USC’s Integrated Media Systems Center, MultEQ XT represents a nonlinear signal-processing approach to the substantial challenge of room response equalization.
Traditional approaches to EQing typically featured time or frequency domain measurements of the room response made at a single (or multiple) listening position(s) and deriving from the results a filter (EQ) response that’s an inverse of the room response. Mathematically, the traditional approach looks like:
heq (n) = filter (EQ) response
h (n) = room response
δ (n) = 1 (for perfect equalization); Kronecker delta function
Or looked at another way (RMS average):
Seems simple enough, but given the complex nature of room acoustics, the traditional approach is a dicey proposition that frequently leads to disappointing results: often requiring considerable subsequent tweaking by ear before arriving at a compromise solution. The “solution” may nevertheless work only for one listening position, if at all. When considered that a linear, electronic device, operating in the frequency domain (your EQ) is being asked to provide a solution to what is largely a non-linear, time-domain problem of an acoustical nature, uncertain, disappointing or otherwise inconsistent outcomes aren’t entirely surprising.
The approach Audyssey has taken - embodied in the MultEQ XT system - is to consider this formidable equalization problem as one of pattern recognition. (Now there is an original approach!) Through the application of a Fuzzy-c means clustering algorithm applied to the raw data gathered during measurement, an equalization solution is found by way of a three-step process: (1) Prototypical representations of room responses are derived from collections of room measurements grouped in clusters, clustered according to their degree (or lack of) similarity; (2) the prototypical responses are further combined, forming a general point response; and (3) the minimum phase component of the general point response is inverted resulting in the required equalization filter response. The process is driven by a target function, which is a pre-determined “ideal” amplitude response curve to which the MultEQ XT attempts to design a best-fit equalization solution.
Editorial Note on Fuzzy-c means Clustering
Clustering, generally speaking, is the process of grouping raw data (in this case, measured room responses) into homogenous clusters having centroids (analogous to a center of mass) or prototypes. Fuzzy-c means clustering is an overlapping, non-hierarchical clustering approach that allows one piece of data to belong to more than one cluster. (The term “fuzzy” relates to how the raw data are bound to any particular cluster by a continuous membership function). Fuzzy clustering assigns degrees of membership of the various measured room responses to appropriate clusters by way of the previously mentioned continuous membership function. The degree of similarity between measured room responses determines the degree of membership held by any particular data in any particular cluster. Mathematically, the Fuzzy c-means algorithm for determining the cluster centroids looks like:
dik = the Euclidean distance between the room responses at positions i and k, respectively.
The goal here is to create a magnitude response, minimum-phase, equalization filter based on a final room response prototype. This final prototype response is actually a linear combination of all cluster centroids under consideration. This step in the process is accomplished by the application of a non-uniform weighting model:
The level of activation of any particular prototype will depend on the degrees of assignment of the measured room responses to the cluster containing the prototype. The final prototype is formed by a nonuniform weighting of the cluster membership function such that the more dense a cluster (in terms of the fuzzy membership function) the greater the contribution will be made by the corresponding centroid in forming the final prototype, with a parallel increase in the effect had on the final multiple-position EQ filter.
Mathematically speaking, MultEQ XT certainly looks good. Now let’s put the system through its paces and see what the amplitude response measurements have to show.
Audyssey Set-Up and Calibration Process
Setting up the MultEQ XT system was fairly straightforward and took, in this particular case, about 10 minutes to complete.
The process begins with placement of the Denon mic at the main listening position, M1, (see Figure 1), positioned at a height approximating the ear height of a seated listener. MultEQ XT depends, by design, on the first measurement run being made at the main listening position. Therefore it is critical the process commences with the Denon mic located at no spot other than M1, wherever that may be in your listening room. A series of repeated test signals are then generated from which the processor gathers its initial calibration data, such as number of speakers, speaker type (sub or satellite), component polarity, and optimal crossover frequency. It then calculates individual speaker-to-mic distances, setting delay & trim for each channel. If the MultEQ XT discovers any problems at this stage, it will let you know that, too. For example, MultEQ XT might find a component opposite in absolute polarity to all the rest. This might be a case of reversed leads or an intended design feature (not uncommon in 3-way loudspeaker systems). Incidentally, MultEQ XT doesn’t care what the polarity of any particular component might be; it merely presents its findings for your use. It has no effect on subsequent calculations.
Once this initial stage is complete, the end-user then moves the measurement microphone to the remaining 5 positions, M2 – M6. At each step a series of test signals are again generated; measurements are made and when completed the mic is moved to the next position, the process repeated until measurements have been completed at position M6. The raw time domain data generated by the collection of measurements provide the building blocks from which the patterns mentioned in the above editorial note are identified. Subsequently an equalization solution, based on a target function, is found and stored. At that point the little green light on the front of the AVR’s faceplate goes on indicating MultEQ XT is active and its all systems go!
Figure 1 illustrates the general layout of the listening room where the measurements were made along with the disposition of the already-mentioned 6 measurement points required by the MultEQ XT system, as implemented in the Denon AVR-3806. (The blue block lines are theater drapes). If you have theatre-style seating and the chair backs can be dropped, the recommended practice is to drop the backs, place the Denon mic, as mentioned earlier, at a height equivalent to listener ear height (or the default 1m) over each seat and measure away. In this case, because the couch back couldn’t be dropped, rather than hang the mic(s) over the couch, I moved the couch out of the way, did the MultEQ XT setup, then moved the couch back into the area for which MultEQ XT had been set up.
The placement pattern was dimensioned approximately one couch wide and two seat rows deep. Try not to cluster the measurement spots too closely otherwise you run the risk of providing MultEQ XT with insufficient low frequency information. The one exception to this recommended practice might be the case where you’re dealing with a highly directional loudspeaker system(s) or a system exhibiting extremely pronounced off-axis lobing. In either case MultEQ might interpret the incoming signal as being indicative of diminished HF content and jack the gain in that part of the audible spectrum, resulting in a system with a hot high end.
Figure 1: Room layout and measurement mic disposition. Note for this series of tests the subs were place directly beneath the front L & R speakers.
Figure 2: Mic positions determined by laser pointers (note red dots).
A series of before/after measurements were made at each of the 6 measurement positions. To be certain that the measurement mics always returned to the same point in space occupied by the Denon-supplied setup mic, a pair of laser pointers were used for positioning .For all measurements, the mics were positioned .83m (~33”) from the floor (average ear height for listeners seated on the couch) . Note that the Denon-supplied calibration mic seen at left in Figure 2 is pointed upward. The mic calibration file hard-coded in the AVR was derived with the mic oriented vertically and requires the mic be positioned likewise when in use. Doing otherwise (eg: pointing it directly at the center speaker) increases the risk of bogus results. Because each calibration is unique to the mic from which it was derived, it is essential that no mic other than the one that shipped with the AVR be used: doing so, once again, increases the risk of bogus results.
Note too that a mic stand was used to position both the Denon and measurement mic. Using a mic stand or at the very least a tripod is the correct approach. Placing the Denon mic on the seat or back of chair/couch or simply holding the thing in your hand can significantly alter the mic’s perception of the incoming acoustic signal, likely causing – you guessed it – bogus results.
Editorial Note on Measurement Mics
If you’re interested in measuring for yourself the effects MultEQ XT has on your system, a caution is in order regarding the mic you make the measurements with.
The LinearX M31 mic used to record the various amplitude response plots featured in this article is oriented vertically, pointing upward towards the ceiling. This is a commonly recommended orientation for the sort of measurements done here. However, the calibration file supplied with the M31 assumes a free-field frequency response to sound waves that are at normal incidence (perpendicular) to the mic’s diaphragm. That is to say, the validity of the calibration is based on the assumption that the mic is pointed directly at the loudspeaker(s). With the microphone rotated 90°, that assumption is no longer entirely valid and a second calibration file is required to maintain accuracy. This second calibration file assumes a pressure-field frequency response to sound waves that are now parallel to the mic’s diaphragm. Just how and where the M31’s frequency response is affected by this change in orientation can be seen by looking at the differences in the required calibration curves, illustrated in the figure below.
M31 free-field (blue) and pressure-field (red) calibration curves.
The two separate calibration curves needed in order to maintain measurement accuracy doing free- or pressure-field measurements can be seen to diverge, beginning ~ 2.5kHz. In actual use, orienting the M31 vertically without using the red pressure-field calibration curve results in response plots that appear to roll off at the HF end of the audible spectrum much more quickly than they actually are, effectively rendering the measurements inaccurate. Herein lies the caution: where required, use the correct calibration file for the type of measurement you are performing!
Confused about what AV Gear to buy or how to set it up? Join our Exclusive Audioholics E-Book Membership Program!