Virtually all high quality musical program material available today was either recorded in stereo or recorded on more than two channels but mixed down to stereo for distribution on compact disks or vinyl records. Given that the object of a high-end audio playback system is to reproduce the original performance as closely as possible, it is obvious that providing surround capability would further that goal immeasurably. In a live performance sound arrives from all directions creating a sense of envelopment and clueing the audience into the unique acoustics of the performance space. Stereo playback can only offer a glimpse of these qualities. Yet, past attempts to add surround sound to stereo playback systems have been mostly disappointing. The reasons for this can be summarized as follows. The difference signal between the left and right channels is used to recover ambience information from stereo recordings. This L-R signal, often called the surround signal, and is then used to drive a pair of speakers located behind the listening area. There is also usually a delay of between 15 and 30 milliseconds inserted into the signal paths of both rear loudspeakers. Because of the precedence effect this delay prevents sound arriving from the rear pair of speakers from interfering with the front stereo image. So far so good. But there are additional factors that must be addressed for the ambience to be accurately reproduced in normal listening rooms.
First, the rear speakers must be decorrelated. This means that sound arriving from the left-rear speaker should not be strongly correlated with sound arriving from the right-rear speaker. The human hearing system will interpret sounds as possessing spaciousness only when they are not strongly correlated between the left and right ears. More precisely, spaciousness is experienced by a human listener when the interaural cross-correlation is small. Direct sound, in contrast, is strongly correlated between the left and right ears. In a real hall the decorrelation of rear arriving sounds is a natural result of the acoustical properties of the hall itself. Although the L-R surround signal does indeed contain ambience information about the original recording space, it is a monophonic signal and will not be interpreted as ambience if presented to both ears monophonically. The standard solution to this problem is to drive the rear speakers 180 degrees out of phase. Although this method provides the necessary decorrelation it also results in severe bass cancellation in the surround sound field.
Second, the in-room frequency response of each rear speaker needs to be well matched to its corresponding front speaker. This means that the in-room frequency response of the left-rear speaker match the in-room frequency response of the left-front speaker. The right-rear and right-front speakers must be similarly matched. One of the clues to the character of a performance space is how the sound of musical instruments is modified by that space. In a live performance listeners do this by comparing the direct sound made by an instrument or ensemble of instruments with the sound of the reverberant tail heard between the notes. This comparison leads them to characterize a hall as "warm", "live" or "bright" sounding. In any case, if the in-room frequency response of the front and rear speakers in the playback system are not well matched then this direct-to-reverb comparison will be colored by the characteristics of the listening room. Such coloration is obviously undesirable. Using well matched loudspeakers, while helpful, is not sufficient because the effects of the listening room normally dominate the in-room frequency response of loudspeakers. Therefore, equalization of the rear speakers is usually necessary. Equalization of the front speakers is usually not required because the ear is much more sensitive to small mismatches between the front and rear speaker's in-room frequency response as compared to any small absolute deviations from a flat in-room frequency response of the front speakers.
A high-end surround stereo system also needs to accurately reproduce bass bloom. Bass bloom is reverberation at bass frequencies. It is usually experienced as a tactile sensation of bass energy that seems to wash over the listener like a beach wave. Reproduction of this experience in a listening room requires the use of full range rear speakers with deep bass response as well as a method of decorrelating them that does not cause bass cancellation.
All of these requirements can be satisfied by employing a suitable dual-channel digital parametric equalizer as a surround stereo processor. One such commercially available and reasonably priced parametric equalizer is the Yamaha YDP2006. This unit uses high-quality 20-bit A/D and D/A converters and is of very high sound quality. Other key features include independently adjustable delays for each of its two channels as well as a precision attenuator in each channel for making level adjustments.
Adding a delay to the rear or surround channels is common. It is part of every Dolby Pro Logic decoder and has a very good justification. A delay of between 15 and 30 ms prevents the ear from confusing first arrivals from the rear speakers with those coming from the front speakers. Such confusion will have the effect of distorting the stereo image. By adding sufficient delay to the rear channels the ear ignores them as far as its processing of transient information thus preserving the normal stereo image. Normally, identical delays are inserted into both the left-rear and right-rear speakers.
A simple but very effective method of decorrelation is to insert somewhat different delays into the signal paths of the left-rear and right-rear speakers. Consider that in a real hall (except for a few special seating locations) the reflected sound arriving from the left-rear is always delayed by different time than sound arriving from the right-rear, both delays being relative to the delay for sound that travels directly to the listener from the stage. Fortunately for this purpose, the Yamaha YDP2006 equalizer provides for independently programmable delays in each of its two channels. The purpose of independently programmable delays in a dual-channel digital equalizer is to provide distance correction for speakers used in large sound reinforcement systems. Nevertheless, the same feature can also be used to decorrelate the rear speakers used in a surround stereo system.
When a significant delay difference is introduced into the rear speakers they are indeed decorrelated but a comb filter effect is also introduced into the frequency response of the surround sound field at the central listening position. Comb filtering creates nulls in the in-room frequency response at odd multiples of the first or lowest frequency null which occurs at a frequency of 1/(2d) kHz where d is the delay difference in milliseconds. For a 5 ms delay difference, for example, the first null occurs at a frequency of 100 Hz, the second null at 300 Hz, the next at 500 Hz etc. In general, comb filtering is desirable in the surround sound field because that is what also occurs in the reverberant field of a real hall. However, if the delay difference between the rear speakers is increased beyond 5 ms the first null will move down into the bass region and attenuate the overall bass response. A solution that permits a somewhat larger delay difference is to measure the in-room frequency response of the rear speakers with MLSSA while driving them simultaneously. This allows observation of the comb filtering effects in the bass. Then, the YDP2006 parametric equalizer can be adjusted to eliminate the null. This procedure requires independent adjustments of the left-rear and right-rear channel bass equalization in order to undo the phase cancellation induced by the delay difference. Details on how to perform this procedure are covered later.
Psychoacoustic data suggest that delays longer than about 30 ms will be perceived as reflections and thus tend to create artificial ambience not present in the original performance space. Critical listeners can perceive delays as short as 25 ms as discrete reflections. Critical listeners can also perceive delays shorter than 15 ms as an image shift, that is, their ears integrate rear arriving transients with those arriving from the front thus distorting the stereo image by pulling images towards the side having the lower rear speaker delay. Therefore the maximum usable delay difference between the rear speakers in a high-end surround stereo system is about 8 ms which provides good decorrelation above the first comb filter null frequency of 62.5 Hz.
The psychoacoustic data therefore recommend a 24 ms delay in the left-rear channel and a 16 ms delay in the right-rear channel for an average delay of 20 ms. These delays are all measured relative to the front stereo pair. Additional delay must also be added or subtracted from them to compensate for any difference between the rear speakers' distance from the central listening position relative to the front speakers' distance from the central listening position. Also you must account for the minimum delay added by the digital equalizer. The Yamaha YDP2006, for example, inserts 3.3 ms of extra delay relative to the delay you select using its delay control meaning that you must remember to add 3.3 ms to the displayed delay to get the actual delay. To avoid these uncertainties MLSSA should be used to measure the rear speaker delays acoustically.
A surround sound processor is unnecessary for generating the L-R surround signal. The balanced inputs of the Yamaha YDP2006 can be used to subtract the left and right stereo signals thus forming the L-R signal simply by rewiring a commonly available two-female-to-one-male XLR Y adapter. If you have balanced outputs from your preamp you simply wire the + lead of one female XLR (from the left channel) to the + lead of the male output XLR. Next you wire the + lead of the other female XLR connector (from the right channel) to the - lead of the male output XLR. You will also need a one-female-to-two-male XLR Y adapter to split this signal to feed the two balanced inputs of the YDP2006.
If you have unbalanced preamp outputs you'll have to cut off the two female XLR connectors of the first XLR Y adapter and wire up two RCA phono plugs such that one goes to the + lead of the male output XLR and the other to the - lead to yield the L-R surround signal when connected to the equalizer inputs through the second XLR Y adapter.
Only balanced outputs are provided by the YDP2006 but you can still use it with power amplifiers having unbalanced inputs using commonly available balanced-to-unbalanced adapters. These adapters use only the + signal of the balanced outputs so you will lose the superior noise rejection of true balanced connections to your power amplifier.
A high-end surround system should use high quality full range speakers for the rear channels as well as for the front stereo pair. The author's system uses Dunlavy SC-Vs in front and Dunlavy SC-IVs in the rear. Time-coherent speakers such as the Dunlavy's are preferable at least for the front stereo pair. In any case, it is recommended that all four speakers be from the same manufacturer and all be of similar design. Be sure to provide enough distance between the rear speakers and the central listening position to insure complete integration of all the drivers. Large speakers with many drivers will require more distance than smaller speakers with fewer drivers.
Experience has shown that certain speaker setups lead to a greater sense of spaciousness than others. The front speakers, for example, should be positioned to subtend a total angle of 60 degrees when measured from the central listening position. The rear speakers should be positioned to subtend a larger total angle, at least 90 degrees but less than 140 degrees. In many listening rooms it is difficult to position the rear speakers any wider than 90 degrees apart without running out of space or moving them too close to the listening area. In the author's system the front speakers are located 10 feet from the central listening position and subtend a total angle of 60 degrees. The rear speakers are located 9 feet from the central listening position and subtend a total angle of 90 degrees. Wider rear speaker spacings of up to 140 degrees can be tried if room dimensions permit.
The rear speakers require equalization especially in the lower midrange because in most setups there will be a sofa placed at the desired listening position. The back of the sofa modifies the in-room frequency response of the rear speakers relative to the front stereo pair. The Yamaha YDP2006 provides the necessary equalization of the rear speakers. MLSSA in its Adaptive Window measuring mode should be used to perform all in-room frequency response measurements.
The need for correct equalization and level adjustments of each rear speaker to match its corresponding front speaker cannot be overemphasized. Only when the in-room frequency response of each rear speaker is well matched to that of the corresponding front speaker will you hear a highly realistic impression of the original acoustic space combined with a clear, deep and panoramic stereo image.
Use of the same power amplifier or at least the same power amplifier design for all four channels is recommended. Initial experiments using different power amplifiers in the rear and front channels quickly revealed that small differences in their sonic signatures were clearly audible. In effect, the unique sonic signature of each power amplifier can act as a clue allowing a critical listener to distinguish between sound originating from the rear speakers vs. the front speakers.
Generally, the rear speakers require about half as much power as the front stereo pair. This was determined by measuring the peak and RMS values of the L, R and L-R signals with MLSSA during peak passages when playing back various recordings. Note that there may be exceptions to these estimates.
Transparent power amplifiers are preferable to those that impart a sweet or euphonic coloration to the sound. Much of the dislike for accurate power amplifiers by some listeners stems from the stereo playback system itself in which virtually all the sound arrives at the listener from the front in rooms with little reverberation. This frontal sonic assault can add an edge to the sound not heard in live performances and there is sometimes an etched quality to the stereo image which can sound overly focused. In a stereo surround system, in contrast, sound arrives from the rear as well as the front largely eliminating these artifacts of stereo playback systems.
Any amplifier can be used in a stereo surround system, of course, but the softer sounding ones might end up clouding up the stereo image or reduce spaciousness. If that happens and you are using accurate speakers, have properly equalized the rear speakers and have correctly adjusted all sound levels then try a more accurate power amplifier.
All power amplifiers used in the author's system are Pass Labs models Aleph 0 and Aleph 0s. The Aleph 0s is the stereo version of the Aleph 0 monoblock. The front stereo pair of SC-Vs are driven by Aleph 0 monoblocks while the rear pair of SC-IVs are driven by a single stereo Aleph 0s.
MLSSA should be used to make the delay adjustments of all speakers including the front stereo pair.
Place an omnidirectional measurement microphone at the central listening position and point it towards the ceiling. The central listening position is defined here as the position midway between the ears of a typical listener when seated comfortably midway between the front stereo pair of speakers.
Start MLSSA with setup file ADAPTIVE.SET. Drive both stereo input channels together and in phase to measure the combined front speakers' impulse response. If the arrivals are coincident you will see a single main spike as illustrated by the solid curve of the figure below. Otherwise, you will be able to distinguish two separate arrivals as illustrated by the dotted curve. If the arrivals are not coincident you will need to adjust the right-front or left-front speakers' distance to the microphone as required to achieve coincident arrivals.
Next, measure the impulse response again this time driving only the right stereo input channel and with both rear speakers also active. The result should look like the figure below which shows the desired rear channel delay of about 16 ms for the right-rear speaker and about 24 ms for the left-rear speaker both relative to the first arrival from the right-front speaker shown at the extreme left. These rear channel delays are not critical and are adjusted using the variable delay feature of the YDP2006.
All equalization and level adjustments should be performed with MLSSA in its Adaptive Window measuring mode. It's not necessary or desirable to equalize every wiggle in the measured in-room frequency response. Rather, the goal of equalization should be to correct for only relatively broadband deviations from the desired in-room frequency response.
Use an omnidirectional measurement microphone placed at the central listening position and point it directly towards the ceiling. The measurement microphone should be pointed vertically for all in-room frequency response measurements to avoid any left/right or front/back biases and in order to give equal weight to all lateral room reflections. Smaller diameter microphones (1/4 inch) are preferable for making in-room frequency response measurements because they are less directional than larger diameter microphones at high frequencies. If you are using a microphone having a diameter of 1/2 inch or more it's advisable to carefully adjust it to point vertically using a plumb line or carpenter's level as a guide.
Start MLSSA with setup file ADAPTIVE.SET.
The figure below shows the properly windowed impulse response of the right-front speaker. The marker is placed just before the first arrival and the cursor is placed 50 ms beyond it to form a time window 50 ms wide. Also shown are two dotted curves generated by the FFT Window Display command. The shorter one illustrates the shape of the time window that MLSSA's Adaptive Window feature uses for computing the in-room frequency response in the midrange and treble regions. Starting at about 120 Hz however the adaptive time window automatically expands to include more and more room reflections until virtually all room reflections are included in the deep bass region as shown by the longer dotted curve which extends beyond the edge of the graph.
Note that RTAs have no time windowing capability whatsoever. In effect, they use a time window that is essentially infinite at all frequencies. RTA measurements can be misleading in the midrange and treble because psychoacoustic data has shown that the ear has an integration time of about 50 ms in those regions when listening to music. Later arriving reflections are perceived separately as reverberation or echoes and therefore do not contribute to the perceived spectral balance of the direct sound. In the bass region however adaptively windowed and RTA measurements yield essentially identical results. This explains why RTA measurements are often used successfully for bass equalization but are not considered very reliable by many sound engineers at higher frequencies.
Spatial averaging of in-room frequency response measurements is not recommended in a high-end surround stereo system since the objective is to achieve a lifelike sound field at the central listening position even if the off-center positions suffer a little. Nevertheless, the sound quality is still quite acceptable for off-center listening positions if the equalization and level adjustments are performed properly.
Parametric equalizers such as the Yamaha YDP2006 are recommended over graphic equalizers because they are more versatile and allow a progressive equalization procedure: Start by inserting a single parametric section with a relatively low Q (broad bandwidth) and adjust that section's center frequency, level and Q (or bandwidth control on some models) to correct the broadest band deviation from the desired response. Next, insert a second parametric section and adjust it to correct the next narrower-band deviation etc. It's time to stop when you've either achieve the desired response or run out of new parametric sections to insert into the signal path. The Yamaha YDP2006 parametric equalizer allows up to six parametric sections per channel. Use as few parametric sections as possible each with the lowest Q setting possible to achieve the desired in-room frequency response.
The figure above shows the unequalized in-room frequency response of the right-rear SC-IV (solid) and the right-front SC-V for reference (dotted) both measured with an Earthworks M30 1/4 inch measuring microphone located at the central listening position. Note the bump in the midrange of the right-rear SC-IV response relative to the right-front SC-V response. This deviation is due to the sofa.
You should match the in-room frequency response of the right-rear speaker to the right-front speaker in order to follow its overall shape without trying to correct for narrow peaks or dips. This is an iterative process that requires making both level and equalization adjustments. At this stage you should not attempt to correct the bass response which is handled by a separate procedure after both rear speakers are equalized in the midrange and treble.
Start by measuring the right-front speaker's in-room frequency response making sure that all other speakers a silent. Store this result as an overlay using the Overlay Store command. You can then use the Overlay Plot command later to plot the right-front speaker's response to compare it to the right-rear speaker's response. Execute Overlay Plot followed by Calculate Level and press TAB to select the Dolby frequency range of 500 Hz to 2 kHz. MLSSA will display the midrange level of the right-front speaker in units of dB-SPL/200 mV and also correct for any overlay offset you previously entered (see next paragraph). You should make a note of this level for future reference.
To measure the right-rear speaker drive both stereo input channels out of phase and disable both front speakers as well as the left-rear speaker. The author uses a Rane FB44 Balance Buddy (now obsolete) to drive the stereo inputs out of phase using its handy phase inversion switch but it may be easier for you to drive only the left or right input channel when measuring the rear speakers. This will result in a desired rear speaker level that is exactly 6.02 dB lower than the corresponding front speaker level. You can correct for this in MLSSA by executing the Overlay Offset command and entering -6.02 dB. That way when you overlay the front reference speaker's response using the Overlay Plot command it will also be displayed 6.02 dB lower in level. Be sure to measure each front and rear speaker in isolation by either disconnecting the other speakers from their power amplifiers or unplugging their power amplifier signal feeds.
It's convenient to create a simple measurement macro by pressing Ctrl-r to turn on macro record followed by Go Once, Overlay Plot and Macro Repeat. Press Ctrl-r a second time to turn off macro record. Execute this macro through the Macro Execute command to continually re-measure the right-rear speaker's in-room frequency response and also overlay the stored right-front speaker's response for comparison. Adjust the right-rear level first to get a good fit in the upper midrange and treble. Next, start making equalizer adjustments to match the midrange. You may also have to make some additional level adjustments to achieve a good overall match.
To make the rear channel level adjustments use the YDP2006's digital attenuator feature which offers digital level control in 0.1 dB steps. It is recommended that you set the ganged analog level control of the YDP2006 to the zero dB position and make all the level adjustments using the digital attenuator. If your rear and front speakers are driven with power amplifiers of equal gain and are about the same distance from the listening position then the amount of attenuation required will be about 6 dB when the analog level control is set to the zero dB position.
When you have achieved a good match by eyeballing the two curves its time to make a precise final level adjustment. Press Esc to stop the macro and execute Go Level Repeat and press TAB to select the Dolby standard frequency range of 500 Hz to 2 kHz. MLSSA will continually re-measure and display the pink-weighted average level in that frequency range. Adjust the right-rear level using the digital attenuator control on the equalizer until it matches the previously noted right-front level within 0.1 dB. If you have to change the attenuator setting by more than 1 dB to match levels this indicates that your frequency response match was not so good. In such cases repeat the whole equalization procedure.
Repeat the entire procedure for the left-rear speaker after first measuring and storing the left-front speaker's measured in-room response as an overlay. Start with the same input attenuator setting for the YDP2006's left-rear channel that you ended up with for its right-rear channel.
The figure above shows the result when the adjustments are done correctly. Here an excellent match is achieved in the midrange where the ear is most sensitive. Note however the large dip at 100 Hz in the bass and the 3 dB mismatch in the high treble over 5 kHz. These deviations will be explained later.
For highest accuracy it's best to position your measurement microphone at the right ear position when adjusting the right-rear channel and at the left ear position when adjusting the left-rear channel. If you do this remember to use the same microphone position for the front speaker reference measurement as you use for the corresponding rear speaker measurement.
The delay difference of 8 ms between the two rear speakers creates a null in the rear speakers' overall bass response at a frequency of 62.5 Hz theoretically. The null frequency will usually be a different in practice due to room effects. To eliminate this null using the Yamaha YDP2006 equalizer you need to measure the in-room frequency response at the central listening position with both rear speakers driven together but with all the other speakers silent. Insert a parametric section of the YDP2006 into the left-rear channel with a Q of 2 and set its center frequency a little below the null frequency to say 50 Hz and select 2 dB of cut. Next, insert another parametric section into the right channel with a Q of 2 set its center frequency slightly higher than the null frequency to say 70 Hz and select 2 dB of cut. Because one section is tuned above the null frequency and the other below it there will be opposite phase shifts introduced into each rear channel at the null frequency which will tend to undo the phase cancellation induced by the 8 ms delay difference. You can also choose to set each section to the same amount of boost. Whether to use a boost or cut setting depends upon whether the overall bass response needs to be cut or boosted. Make fine adjustments to each parametric section's center frequency, Q and gain to eliminate the null while maintaining as far as possible a flat overall rear-channel bass response.
This procedure takes some practice but can be done as illustrated by the figure above. The dotted curve clearly shows the comb filtering null at 55 Hz while the solid curve was obtained after proper adjustment of the equalizer to obtain a flat response in the bass.
When performing the bass measurements it is recommended that you disable smoothing through the View Magnitude Decibels command as shown in the figure above. This allows you to easily identify any room-induced nulls in the bass region in order to avoid inadvertently trying to correct them by equalization. You should not attempt to correct any such room-induced nulls. Doing so could result in overdriving your speakers or power amplifiers or you could add a huge peak to off-center listening locations. In the figure above there is clearly visible a room-induced null at 100 Hz which is narrow enough to be of no concern and should not be corrected by equalization. The large number of higher frequency nulls are due to comb filtering effects. Because these are not in the bass region they are desirable. Note also that these higher frequency comb filtering nulls are visible in this plot only because there is no smoothing. They would not be noticed if smoothing were enabled as in the previous plots.
The ganged analog level control on the YDP2006 can be used to slightly increase or decrease the level of both rear channels by the same amount to suit your taste. A slightly higher surround level (+0.2 dB higher level in both rear speakers) deepens the soundstage and narrows the ambience while a slightly lower surround level (-0.2 dB) does the opposite.
As a final note be aware that any equalizer adjustments you make to a speaker can affect its output level. Therefore, whenever you make any equalizer adjustments remember to re-measure that speaker's level using the Go Level Repeat command and re-adjust it if necessary.
The L-R surround signal is presumed to carry mostly ambience information because for direct sound arriving at a pair of coincident recording microphones the phase and amplitude differences between the L and R signals are usually small resulting in their mutual phase cancellation when subtracted. Hall ambience is usually acquired by a pair of widely spaced recording microphones placed further away from the performers than the main stereo microphone pair. The phase relationship between the L and R signals for widely spaced microphones is essentially random allowing the ambient information to appear in the L-R surround signal.
Coincident stereo microphones are not truly coincident. They are normally stacked one on top of the other or sometimes intentionally moved apart laterally by a few inches to improve stereo imaging. Considering that at 10 kHz the wavelength of sound is merely an inch long, any direct sound arriving at an angle will not reach each microphone simultaneously resulting in a relative phase shift between them that can be very large at high frequencies. A large phase difference between the L and R signals will cause an excessive amount of direct sound to leak into the L-R surround channel. Such leakage is undesirable because it essentially converts what is really direct sound into ambient sound in the playback system and can make the overall spectral balance tend towards brightness on some recordings.
The solution is to shelve down the treble response of both rear speakers by 2 or 3 dB above 5 kHz using two parametric sections of the YDP2006 for each rear speaker. You should also add a low pass filter section of the YDP2006 to each rear channel tuned to 16 kHz to remove all energy above that frequency. These final adjustments explain the 3 dB mismatch above 5 kHz previously noted between the right-rear and right-front speakers' in-room frequency response curves.
The author's system also includes a Velodyne FSR-18 subwoofer crossed over at 40 Hz from the front Dunlavy SC-Vs. Without the subwoofer there is a deep null at 25 Hz due to the front speaker positions relative to the room when they are set up for optimum stereo imaging. Placing the subwoofer between the two rear speakers where there is no room-induced null solved the problem and extended the front channel bass response down to 16 Hz (-3 dB).
The front Dunlavy SC-Vs are bi-amped such that a Klark-Teknik DN410 stereo parametric equalizer can be inserted into the SC-V woofers' signal paths in order to equalize the mid and upper bass room modes. Bass equalization of the front speakers may not be necessary in all setups. If you do insert an equalizer into the woofers' signal paths do not use a digital equalizer as all of them insert a minimum delay which is high enough to completely destroy the time coherence between the woofer and higher frequency drivers. Instead use a good analog parametric equalizer.
