Images and measurements of my current audio system

A few pictures and descriptions of the equipment. March 04

(click image to see larger version)

(click image to see larger version)
This slide shows a schematic illustration of how the audio system is configured, how the various digital and analogue inputs are handled. Input to the DCX2496 is full scale, so that the internal representation makes use of the full dynamic range of the Sharc DSP, how the full scale balanced output of the Behringer is fed through analogue volume controls, which also act as phase splitters to bridge the tripaths.

(click image to see larger version) Details of the midrange and treble speakers, the full unit is shown on the right, with the backs of the transducers shown on the left. The TAD compression driver is mated to the back of the oak tractrix horn. Note that the magnet of the JBL is the same diameter as the 10" cone!

(click image to see larger version) The business end of the Altec 18" woofer, and the innards of the central portion of the 4 meter transmission line. Note the magnet of the Altec is quite small, this is so the TS parameter of Qts will be relatively high. The woofer is intended to be used in a large volume box, for low frequency response. (And yes, that is zero gauge arc welders cable that I am using for the internal wiring).

(click image to see larger version) Details of the equipment rack (left) and rear wiring (right). The tripath and volume control cases are rack mount, but open to the back. The whole thing is a bit of a wiring tangle, but it works.

(click image to see larger version) Close up of the "messy" side of the Tripath rig. This comprises four stereo TA104a evaluation boards (on the right) with two separate power supplies, one for the right speaker stack, and one for the left speaker stack. Each power supply has a 2.5kW torroid transformer (hidden behind the wires), and runs at + / - 87v. The four 47,000uf filter capacitors are seen along the top of the unit. The storage capacity has enough Joules of energy to lift my BMW car a half inch off the ground!

(click image to see larger version) Detail of the tripath TA2021B treble amp, and the ganged volume control. I have plans to simplify this portion of the system.

(click image to see larger version) The Tripath amplifier rig at various stages of testing and construction. Clockwise from top left: A breadboard test version with a single tripath eval board. I used this simple rig to evaluate and listen to the tripath on a variety of speakers. Top right: showing the pair of torroidal 2.5kW transformers, power supply capacitors, and four tripath evaluation boards bolted onto the 19" rack mount steel front plate. It was a tight squeeze, but everything fits. Bottom right: The four eval boards, bolted together, with speaker banana plug leads mounted on Plexiglas. Bottom left: the completed unit, sitting on the floor of my workroom. The additional power supplies, and wire connectors are now mounted on a couple of Plexiglas plates, covering the torroid transformers.
The images don't convey the physical size of this beast, or the weight.(it's heavier than my spouse) The unit packs 5 kW of output power capability (8 channels) into about 1.5 Cu.Ft of volume. This is only possible because of the extremely high efficiency of the tripath modules, which run better than 80% efficient. It would be impossible to dissipate the heat generated by a 5kW Class A/B amplifier in a compact 1.5 Cu.Ft unit such as this. Also note, that it is futile to try to get 5 kW out of a U.K. wall socket. Even at 240 Volts x 13 Amps, this is only 3120 W. I built the unit with two separate power cords, so that I -could- plug each one into a separate circuit in the house. Of course I don't, bother, because I really only need the peak voltage swing from the amps, not the continuous rating.

Acoustic measurements of the sound system in the current listening room. May 04

Just hot off the press! These measurements (without comment for the time being), were made just two days ago (26th May 04) after I re-worked the midrange and horn EQ to make the system more balanced:

(click image to see larger version) Right and left channel 1/3 octave smoothed frequency response, 20hz to 20khz, log scale, 30db total vertical scale, (right top, left bottom.) FFT time window is 200ms, and includes both the direct arrival, and reverberant wave field of the room. Frequency response of both channels is essentially bounded between troughs of 94 dB and peaks of 100dB, making the system flat with only + - 3db variation across the entire band, creating a relatively even tonal balance.

(click image to see larger version) Right and Left channel impulse responses calculated from the ETF MLS method. 3 ms. time scale. distance at 11ft.(both ear height)

(click image to see larger version) Right and Left channel impulse responses calculated from the ETF MLS method. 20 ms. time scale. distance at 11ft.(both ear height) The low level of energy following the impulse indicates that the side walls, floor, and rear wall do not present any significant reflections that will cause problems for imaging. (This prediction of good imaging characteristic is borne out when listening to recorded music) The reflections between 3-5ms are likely due to the bare ceiling however treatment of this area is problematic, and may affect the sense of image height.

(click image to see larger version) Right and Left channel unsmoothed low frequency response, 2hz to 200hz, linear scale, 30db vertical scale, measured with the mic at distance, 11ft. The FFT window is 460ms long, four graphs per chart, each initiated 40ms later than the previous.
Low frequency response is flat down to an amazingly low 8Hz

(click image to see larger version) These 3D waterfall plots (spectral analysis as a function of time) illustrate the low frequency behavior of the woofers in the listening room. Frequency scale runs from 2 Hz to 200Hz,(linear) and dB scale runs from 100 dB down to 70 dB, over a temporal analysis range of 300ms with an FFT gate time of 90ms),

(click image to see larger version) Right and left channel 1/12 octave smoothed frequency response, 20hz to 20khz, log scale, 30db total vertical scale, (right top, left bottom.) FFT time window is 200ms, and includes both the direct arrival, and reverberant wave field of the room. Frequency response of both channels is essentially bounded between troughs of 82 dB and peaks of 100dB, indicating the peaks and valleys due to interference from all the room reflections.

(click image to see larger version) Pseudo anechoic frequency response, 1/3 octave smoothed high frequency response, 20hz to 20khz, log scale, vertical scale100db down to 70db, measured with the mic at the distance 11ft. The FFT window length was shortened to 4ms for this plot. The short FFT spectral analysis window was chosen to exclude all of the room reverberation energy, and highlight only the treble direct arrival response This gives a more accurate indication of what the treble sounds like, the curves are ripple free, and very flat from 1khz up to 20khz The 2002 compression driver / horn combination dominates the response of the first arrival such that the response above 1khz is basically flat to within + / - a couple of dB. The directivity of the horns above 2khz, coupled with the distance of each speaker from the side walls, along with the aim of the speaker adjusted toward the center of the rear wall, all conspire to minimize the detrimental effect of side wall reflections on overall imaging of the first arrival,

(click image to see larger version) phase response

(click image to see larger version) These 3D waterfall plots (spectral analysis as a function of time) illustrate the high frequency behavior of the tweeters in the listening room. Frequency scale runs from 20 Hz to 200Hz (linear), and dB scale runs from 100 dB down to 70 dB, over a temporal analysis range of 1.5ms with an FFT gate time of 0.7ms), This display is a good indication of the quality of the treble in the system.

(click image to see larger version) Frequency response of the left channel, generated by taking a 2 second fft of the ETF impulse response, and graphing the result with no smoothing. This is the unsmoothed, non-averaged response of the room and system, and takes all reflections, reverberation, and decay into account.

(click image to see larger version) Left channel impulse response calculated from the ETF MLS method. aprox 2 ms. time scale. distance at 11ft.(both ear height) graph from cooledit showing the actual sample values of the impulse response.

(click image to see larger version) Right channel impulse response calculated from the ETF MLS method. aprox 2 ms. time scale. distance at 11ft.(both ear height) graph from cooledit showing the actual sample values of the impulse response.

(click image to see larger version) Frequency response of the right channel, generated by taking a 2 second fft of the ETF impulse response, and graphing the result with no smoothing. This is the unsmoothed, non-averaged response of the room and system, and takes all reflections, reverberation, and decay into account.

Now, following just below are some measurements made prior to modifying the tonal balance. These are graphs of the first measurements that I made after moving to my new town house, and setting up the audio system for the first time. (March 04) Equipment used: IBM laptop, and Panasonic mic element, with ETF and CoolEdit software. All of the crossover settings, transducer compensations, and level settings were left unchanged from the prior settings in the previous set up in the old house in Wimbledon. Although these graphs can convey a great deal of objective information related to the performance of the system, the question always arises: "how does it sound?", the best answer, as usual, is: Great! but not nearly as good as it is going to sound after I tinker with it a little bit more!

(click image to see larger version)

(click image to see larger version) Left channel impulse responses calculated from the ETF MLS method. 3 ms. time scale. Top graph taken at a distance of just under 10ft from the speaker, middle graph at 11ft.(both ear height) and lower graph at 12ft from the speaker, with the microphone above ear height (4ft) Sorry for the multi-coloured lines, Doug Plumb at ETF got a bit carried away with showing the measured, and soundcard responses. The Black line is the calculated impulse response for the system and room at the microphone position. Interpretation of these impulses shows the polarity to be a bit ambiguous, and the peak to trough ratio slightly too large. This can possibly be adjusted using the fine delay settings on the digital crossover. Of particular note is the low level of energy following the impulse peak and trough, this denotes the desirable absence of early reflections from the rear walls of the cabinets, or diffractions from the cabinet edges.

(click image to see larger version) Left channel impulse responses calculated from the ETF MLS method. 20 ms. time scale, distances of 10,11 and 12ft. from the speaker as before. Again, the low level of energy following the impulse indicates that the side walls, floor, and rear wall do not present any significant reflections that will cause problems for imaging. (This prediction of good imaging characteristic is borne out when listening to recorded music) The reflections between 3-5ms are likely due to the bare ceiling however treatment of this area is problematic, and may affect the sense of image height.

(click image to see larger version) Left channel 1/3 octave smoothed frequency response, 20hz to 20khz, log scale, 30db vertical scale, at three microphone distances (10, 11 and 12ft.) FFT time window is 200ms, and includes both the direct arrival, and reverberant wave field of the room. All the graphs are essentially bounded between troughs of 88dB and peaks of 100dB, with relatively even tonal balance. It is interesting to note that changes of as little as one foot can make a large difference to the mid and low frequency response. This indicates the dimensions of the room yield a very well spaced set of resonance modes, which translates to an impression of unbiased bass response.

(click image to see larger version) Left channel unsmoothed low frequency response, 2hz to 200hz, linear scale, 50db vertical scale, measured with the mic at three distances (10, 11 and 12ft.) The FFT window is 480ms long, with each successive graph initiated 80ms later in time than the previous graph (The top graph, in red, was initiated at the zero time of the impulse) These graphs are similar to a "flat waterfall plot". and show both the direct arrival frequency response, but also the low frequency, undamped modes and reverberations, seen as "stacked peaks" on the graph. The worst case undamped peaks are at the positions of the major axial modes for this room: (30hz, 41hz, and 69hz). These modes are prominent compared to the rest of the frequency band, however the room and furnishings provide sufficient damping such that the overall decay is actually lower than in either of the previous two rooms I have used in London.

These set of graphs best illustrate the amazing response of the 18" Altec woofer system / room combination down to a phenomenally low 8Hz. Interestingly, this is by far the deepest frequency response that I have measured in any room, or with any of the systems I have tested or built.

Most audiophiles are lucky to have speakers which extend down as low as 32Hz. (my expensive M&K subwoofer only has usable response down to 35Hz), yet the big 18" Altec's in the 4 meter transmission lines have a power band response that extends a full two octaves deeper! Also note that the response at 2Hz is only 10-12dB lower than the typical midband SPL, and is no were near the noise floor, which is a good 20dB lower. The pro-sound Altec drivers are capable of SPL levels in excess of 113dB each across the whole of this frequency range, so the above graphs represent a true picture of the bandwidth that is realized in practice, at the listening position, measured with an excitation signal of approximately 95dB for each driver individually. (As opposed to close mic, low amplitude, unrealistic measurement conditions).

The Altec/transmission line speakers have been previously measured by me in two other rooms, using both electronic bass boost and (two different) digital crossovers (BSS and Behringer) with digital bass response correction, yet the response has never previously been lower than about 11Hz.to 15Hz. The current response tailoring is the same as in the previous listening room in Wimbledon. The digital bass response correction is set as a second order shelving filter, starting at 20hz, and the additional low frequency extension below this shelving point must be due to the room / speaker interaction.

Overall bass efficiency is likely due to the proximity of each driver to the three way corner (masonry walls and cement ceiling) such that over their entire bandwidth (below 110Hz) the Altec drivers are radiating into 1/8 space, i.e., the three rigid acoustic boundaries are within 1/8 of a wavelength for frequencies below 140Hz, which makes the apparent efficiency of the drivers about 9dB higher than if they were radiating into free space. (aprox 3dB for each boundary, ignoring absorbs ion and transmission losses.)

Furthermore, the dimensions of the room are such that for all frequencies lower than 35Hz, the shortest dimension, the 8ft ceiling height is less than 1/4 of a wavelength, and for all frequencies below 20Hz, the medium dimension of 14ft. will be less than 1/4 of a wavelength, and finally, for all frequencies lower than 15hz, even the longest dimension of 18.75 feet will be less than 1/4 of a wavelength. As the frequency drops below 35hz, each of the room boundaries will move through from the range of beyond 1/4 wavelength, down to less than 1/8 of a wavelength (at the frequencies of 17.5Hz, 10Hz, and 7Hz respectively). In other words, for frequencies below 140hz, the ceiling is within 1/8th of a wavelength, and augments the sound pressure level by almost 3dB, when the frequencies go below 35Hz, the floor is also close enough to start contributing to the SPL of the sound, and this increases as the frequency is lowered to 17.5Hz, where the floor is 1/8th of a wavelength from the driver, and the phase of the waveform is nearly identical, regardless of height, and the SPL will get almost a 3dB boost below 17Hz. Likewise, for frequencies below 20Hz, the side walls start adding in phase, reaching maximum contribution below 10hz (another 3dB), and the longest dimension contributes below 15Hz, ramping to 7.5Hz at which point the room will be in complete isophase condition.

Of course wave fronts still propagate throughout the room at the speed of sound for these low frequencies, however the rate of change of pressure amplitude will be so slow, and the wavelengths so large, that each point in the room will appear to have identical pressure amplitude, and phase for these low frequency waveforms. At frequencies below 7Hz, essentially all six of the room boundaries will be closer than 1/8th of a wavelength, and this will yield an extra 9dB of gain, for a grand total of 18dB extra SPL when the corner placement is included. The natural progression across the 1/4 to 1/8 wavelength (zero dB to +3dB increase) combined with the staggered dimensions deliver a gradual ramping up of room gain, which mimics the missing boost below 20hz. The trick is particularly effective in the current room, because of the full masonry construction of the four walls, and the uncommon use of prefabricated concrete floor and ceiling, such that the low frequency reflection coefficient is very near unity.

(click image to see larger version) These 3D waterfall plots (spectral analysis as a function of time) illustrate the low frequency behavior of the woofers in the listening room. Frequency scale runs from 2 Hz to 200Hz,(linear) and dB scale runs from 100 dB down to 50 dB, over a temporal analysis range of 400ms with an FFT gate time of 80ms), This type of display also highlights the low frequency, undamped modes and reverberations, seen as high amplitude zones which extend as "ridges" from the back, down to the left of the plot. The worst-case undamped peaks are at the positions of the major axial modes for this room: (30hz, 41hz, and 69hz). While these modes are prominent compared to the rest of the frequency band, they do not dominate the entire range, and the overall wave field is relatively evenly distributed, and damped, with uniform decay over time. The peaks and ridges associated with the rectangular listening space do exist, however the dimensions are such that the modal distribution is extremely regular, with very even mode spacing over frequency, and little evidence of trouble resonances due to stacked or overlapping modes. The room modes that are detectable (particularly the mode at 30 Hz.) can be treated with additional damping (Helmholtz resonator), and/or manual or automatic equalization. More importantly, there are no significant dips in the response (valleys and modal nulls), which would be more difficult to compensate with electronic or digital equalization.

(click image to see larger version) These 3D waterfall plots (spectral analysis as a function of time) illustrate the high frequency behavior of the tweeters in the listening room. Frequency scale runs from 20 Hz to 200Hz (linear), and dB scale runs from 100 dB down to 70 dB, over a temporal analysis range of 1.5ms with an FFT gate time of 0.7ms), This display is a good indication of the quality of the treble in the system, and we expect to see a flat frequency response at zero time (along the back of the waterfall plot) that rapidly and smoothly decays to a low value and that no prominent ridges run from the back to the front. The microphone placement for these three graphs was at 10, 11 and 12ft. from the speakers, putting it back past the center position of the room, in the transition zone between the direct and reverberant sound fields. The effect of this is that for the top two graphs at 10 and 11 ft. the waterfall is very well behaved, with flat response, and rapid decay from 97 dB down to 75dB in less than a quarter of a millisecond, after which the reverberant part of the wave field dominates, and a variety of standing wave patterns (dependant on microphone position) are visible. The bottom graph has the microphone in a part of the room where the reverberant wave field dominates, and the spectra are riddled with numerous ridges and valleys. The noise floor on each of the three plots has ridges and valleys that are not consistent in frequency between the three microphone positions, suggesting that they are caused mostly by room interaction, rather than speaker related problems, i.e.: transducer ringing, horn mouth termination problems, speaker cabinet edge diffractions or internal undamped reflections within the cabinets.

(click image to see larger version) Right channel impulse responses calculated from the ETF MLS method. 3 ms. time scale. Top graph taken at a distance of just under 10ft from the right speaker, middle graph at 11ft.(both ear height) and lower graph at 12ft from the speaker, with the microphone above ear height (4ft) The Black line is the calculated impulse response for the system and room at the microphone position. Interpretation of these impulses shows the polarity to be a bit ambigious, and the peak to trough ratio slightly too large. This can possibly be adjusted using the fine delay settings on the digital crossover. Of particular note is the low level of energy follwing the impulse peak and trough, this denotes the desireable absence of early reflections from the rear walls of the cabinets, or diffractions from the cabinet edges

(click image to see larger version) Right channel impulse responses calculated from the ETF MLS method. 20 ms. time scale, mic placed at distances of 10,11 and 12ft. from the speaker top to bottom graphs. Again, the low level of energy following the impulse indicates that the sidewalls, floor, and rear wall do not present any significant reflections that will cause problems for imaging. The reflections between 3-5ms are likely due to the bare ceiling and may affect the sense of image height. Also of note is the absence of any single strong rear wall reflection, indicating that the reflections from the back of the listening room are low enough in amplitude, and spaced far enough apart in time, to be in the Hass fusion zone, despite the fairly close proximity of the rear wall to the listening position. (Less than 8 feet).

(click image to see larger version) Right channel 1/3 octave smoothed frequency response, 20hz to 20khz, log scale, 30db vertical scale, mic at three distances (10, 11 and 12ft.) FFT window length of 200ms. All the graphs are essentially bounded between troughs of 88dB and peaks of 100dB, with relatively even tonal balance. (+ / - 6dB 8Hz to 21kHz in room response) Of particular note is the response of the 11 foot mic position, which shows less than + / - 3dB over the same frequency band. The graphs of frequency response, for both the left and right channel consistently show a gentle roll off in the treble region, caused by the overall high frequency damping of the room. The horn loaded TAD loudspeakers have frequency response tailoring (via the Behringer DCX2496) that produces nearly flat high frequency response on the direct arrival wave front, however the long FFT spectral analysis time window for these graphs (200ms) includes most of the room reflections, and shows the HF damping effect of the wall to wall carpeting, and furnishings. The high frequency waterfall plots have an FFT window length short enough to represent the anechoic response of the tweeters, (or see the next graph below).

(click image to see larger version) Right channel 1/3 octave smoothed high frequency response, 200hz to 20khz, log scale, vertical scale100db down to 70db, measured with the mic at three distances (10, 11 and 12ft.) The FFT window length was shortened to 18ms for this plot. The short FFT spectral analysis window was chosen to exclude much of the room reverberation energy, and highlight only the midrange and treble direct arrival response and early reflections from the walls and ceiling. This gives a more accurate reflection of what the midrange and treble sound like, the curves are ripple free, and very flat from 400hz up to 20khz The 2002 compression driver / horn combination dominates the response of the first arrival such that the response above 1khz is basically flat to within + / - a couple of dB. The directivity of the horns above 2khz, coupled with the distance of each speaker from the side walls, along with the aim of the speaker adjusted toward the center of the rear wall, all conspire to minimize the detrimental effect of side wall reflections on overall imaging.irst arrival, the response above 1khz is basically flat to within + / - a couple of dB.

(click image to see larger version) Right channel unsmoothed low frequency response, 2hz to 200hz, linear scale, 50db vertical scale, at three distances (10, 11 and 12ft.) The FFT window was 480ms long, with each successive graph initiated 80ms later in time (after the zero time of the impulse). These graphs are similar to a waterfall plot, and show the low frequency response which amazingly extends down to a subterranean level of eight Hertz, and below!

(click image to see larger version)

(click image to see larger version) Audio

(click image to see larger version) RT60 reverb time

(click image to see larger version) Audio

(click image to see larger version) Audio

(click image to see larger version) Audio

(click image to see larger version) Audio

(click image to see larger version) Audio

(click image to see larger version) Audio

(click image to see larger version) Audio

(click image to see larger version) Audio

(click image to see larger version) Audio

BACK