Text preview for : B&W_DM6_research.pdf part of Reprint HiFi News B&W DM6 development Article about designing a phase lineair loudspeaker.



Back to : B&W_DM6_research.pdf | Home

Ph
E audio art seems to THshare of so-called majorhave a n impressive 'break-throughs', with developments in loudspeakers in the vanguard. This is predictable, as there are s o many ways of creating and radiating sound that designers have plenty of scope for their efforts. 1975 saw the release by some four manufacturers throughout the world of 'Linear Phase' dynamic loudspeaker systems, one of which was British. From our recent correspondence files, unquestionably, 'phase linearity' in loudspeakers is interesting, and confusing, many hi-fi devotees. One of the most recent treatments of 'this subject appeared in Wireless World1, in which H. D. Harwood (BBC Research Department) discussed the effects of phase-frequency response in relation to perceived sound quality and stereo image formation. Mr. Harwood's scepticism about the audible effects of non-uniform phase-frequency response is, perhaps, best summarised by a direct quotation from the article: 'Until, therefore, someone can demonstrate that phase effects of this nature affect the sound quality of loudspeakers on programme, and not just o n highly artificial signals, the appropriate treatment appears t o be to pull down the blind and pretend that phase, in this sense, does not exist'. This is also his reaction to stereo image formation effects. This topic has been debated in the audio literature many times before (and latelyWireless World, March 1976: 'Phase and Sound Quality' by James Moir, and Michael Gerzon's Ietter'Audibility of PhaseDistortion')ofcourse, including a reference to it in the early G . A. Briggs' texts2. Other useful material has appeared in B. B. Bauer's article on 'Audibility of Phase Distortion', Wireless World3, and several articles in the JOURNAL OF THE AUDIO ENGINEERING S O C I E T Y . ~1973 E. R . Madsen, In ~ ~ er al, presented a paper t o the 44th Convention of the AES at Rotterdam, on the 'Threshold of Phase Detection by Hearing'=. Another AES paper was published in 1974

& Loudspeakers
Background to the debate by ~ o n a l d Aldous
by Villy Hansen and E. R. Madsen,' dealing amplitude response. This latter shows the with aural phase detection, and last year Bang phase/frequency curve that should be obtained & Olufsen's Erik Baekgaard (assisted by S. H. if the loudspeaker were a completely minimum Pramanik) contributed an important paper on phase system, and comparison of the two the Beovox Uni-Phase loudspeaker systems to curves revealed that the ELS justifies the claim the 50th AES Convention, in L ~ n d o n . ~ 'minimum phase' to a far greater degree than In our advertisement, as well as editorial any moving-coil loudspeaker, including recent pages, we have carried descriptions of the designs claiming particular attention to this B & 0 Uni-Phase loudspeaker designs, aimed feature. at producing a linear phase response. T o obtain The ever-modest Peter Walker, commenting this so-called 'uniformed phase' in the Beovox on these findings, said 'at this point I might family of speakers, a Phase Link is incorpor- retire and state: case proven'. However, he ated. This patented technique covers a goes on, 'an electrostatic loudspeaker differs combination of 12 db/octave filter and an from a moving-coil in so many respects in the additional narrow band loudspeaker driver. way it handles constantly changing waveforms Our investigation of this topic evoked some that if would be premature to ascribe its comments from veteran ELS designer Peter character-or more correctly lack of character Walker (Acoustical Manufacturing Co. Ltd.) -to any one criterion'. He remarks that 'it is possible to set up a At our invitation John Bowers, and his highly artificial source of sound consisting of a technical team at B & W Loudspeakers Ltd., fundamental and a number of harmonics in Worthing, have prepared an article outlining specified phase relationship one t o another. It their design philosophy involved in the two is then possible to change the phase of one of years' development programme for their D M 6 the harmonics t o produce a change in the monitor loudspeaker. This feature covers the perceived sound. In this way it can be shown historical background and some of the prethat there are exceptions to Ohms' famous liminary work in creating a 'minimum phase speaker'. We publish this here as a contribuacoustic law which stated that: aural perception depends only o n the (short time) amplitude tion t o clarifying this subject, but, of course at spectrum of a sound and is independent of the this stage we are not supporting either faction, phase angle of the various frequency com- just trying to help HFN/RR readers. ponents contained in the spectrum'. S e l e c t e d References Mr. Walker goes on t o say that 'in spite of 1. H. D. Harwood: 'Audibility of phase effects i n the above, it would be a brave, and I belieqe loudspeakers', Wireless World, Jan. 1976. foolish, man who would conclude that lack of a 2. G . A . Bri99s: 'Louds~eakers',(First Edition* ~,"~io~,harfedale Wire'ess Idlet minimum phase system in a loudspeaker is, therefore, necessarily detrimental to the 3. B. B. Bauer: 'Audibility of Phase Distortion', Wireless World, Vol. 80, 1974, PP. 27/28. natuwl reproduction of programme material. The harmonic structure of music and other 4. T. Ideal Crossover 'On the Transient AES, 1962 of Robert Ashley: Networks', Journal, natural sounds are randomly distributed intheir 5. A, Schaumberger: Ilmpulse M~~~~~~~~~~ phase relationship and, furthermore, the phases Techniques for Quality Determination i n Hi-F Equipment, with Special Emphasis on are changing in a more or less random manner Loudspeakers', Journal, AES, Feb. 1971. from instant to instant.' 6. E. R. Madsen et al.: 'Threshold of Phase Mr. Raymond Cooke (Manaaing Director, Detection by Hearing', paper at 44th AES Convention, Rotterdam, 1973. K E F Electronics Ltd.) recently decided t o use and E. R. Madsen: 'On Aural 7- Villy his company's computer to produce a curve of Phase Detection' Journal, AES. Jan./Feb. 1974. the phase of a Quad electrostatic loadspeaker, 8. E. Baekgaard: tLdudspeakers: Missing the and also a curve of the Hilbert transform of the. Link', 50th AES Convention, London, 1975.

-

,

~

-.

Some findings and views outlined by John Bowers and Stephen Roe*
far back as middle of the ASlastacousticaltheOhm formucentury, lated an law stating that 'Aural perception depends only on the amplitude spectrum of a sound and is independent of phase angles of the various components contained in the Spectrum'. However this law as stated was soon to be disproved and as far back as the thirties timedelay distortionon telephone lines rendered speech almost unintelligible. The cases disproving the early work by Ohm and Helmholtz are legion. There is the classic case of the AM carrier where the phase on one sideband is changed by 180" producing a quasi-FM signal with an immediate audible difference. M. R. Schroederl in a n excellent paper quotes another:( I ) Take 100 seconds worth of a speech signal ; (2) Fourier transfornl it (assuming 100-2 periodic repetition). This yields a frequency component every 1/100 Hz; (3) Randomise the phase angles, i.e., choose each phase angle from a uniform distribution between 0 and n rad; (4) Inverse Fourier transform. The result is a signal which, for allpractical-parposes,duringintervals shorter than 100 s, looks and sounds like a Gaussian noise (with a power spectrum equal to that of the speech signal). T ~ L I S manipulating the phase angles in this case has not only altered the acoustical quality but completely changed the signal from (intelligible) speech to (random) noise.' Schroeder suggested adding the rider 'short term' before 'anlplitilde spectrum' and both he and Hansen and Madsen? have produced excellent papers quantifying the aural perception of phase effects. In the latter case phase effects as small as 10" were shown to be audible under certain
~ J W Y , , , ,

-

conditions, thus ,,yond any doubt tdar the human being is not 'phase deaf', as previously suggested by Ohmand Helmholtz. The questions that must now be asked are whether waveforms are preserved under normal listening environments, to what degree they are reproduced by the recording process and ancillary equipment driving the loudspeaker, and whether a dramatic reduction in phase distortion which is achieved by !he linear

*B

& W Loudspeakers.

jhase loudspeaker system is u~dible?Let us first examine the basic principles involved.
Time Delay Distortion

Heterodyne ~nalyzer

_ _ _ Sweep_Control_81_ _ - - - _ - Marker Signals

---- 1
I 1 I I

-

Fig. I , with the units mounted on a flat baffle, illustrates the effect of time delay distortion whereas in fig. 2 the units are mounted at their acoustic centres giving a stepped bame configuration. To illustrate time delay distortion we have deliberately simplified the case to a two-unit system where the bass unit is relatively large and the high frequency unit of similar con\ struction but appreciably smaller. : Among other controlling factors the mass of the moving components in the bass unit will be considerably greater than that of the high frequency unit and consequently there will be greater time delay in the transfer from electrical to acoustical energy. Each-drive unit will therefore have an 'acoustic centre' as shown in figs. 1 and 2, which is unlikely to be in the plane of its mounting flange. As a matter of common sense or rather good engineering practice, therefore, there would seem no justification for mounting units with, different acoustic centres on a flat baffle plane, as time'delay distortion is bound t o occur. It may be asked why, if correction of -this distortion can so easily be made, so many loudspeaker systems, some of them very good, have not taken advantage of .the correct mount-

Po

I

I

I
-J
Recorder 2307

I

Loudspeaker under test

ing planes for the drive units. The answer very largely lies in the fact that it is only within the last 3 years that instrumentation enabling phase response of loudspeakers to be easily measured has been available. Whilst all amplitude related measurements are relatively straightforward, phase measurement called for a delay unit which-can compensate

for the transmission delay between the loudspeaker and measuring microphone. A block diagram of equipment used in phase measurement is shown in fig. 3.
Crossover and Filter Networks

f

\

_ _-5

FIG. I

.

i
0

:
j

.

. . . . : . . . . . . .

. . .

:..

. . : .
:

.

Conventional flat baffle arrongemenr

It is well known that phase shift takes place whenever a reactive component is inserted in a network and the object of the loudspeaker crossover and filter network is to divide the frequency spectrum between the various units forming the system in such a manner that their summed output is linear. It is also of importance to know theirsummed phase characteristic in order that arrival times, already linearised by correct unit mounting planes, is not degraded. For simplification we have taken first, second and
47K

third order complementary networks, summed their output and present the results in terms of amplitude, phase and transient performance. Fig. 4 shows the circuit used for this purpose and the results are shown in figs. 5-7. I t will be seen that not only is the first order network the only configuration which satisfies the phase and transient characteristic but it is the only network to satisfy the amplitude characteristic. Other circuit arrangements have been devised which satisfy all three criteria, but these reinforce the case already shown that if absolute amplitude linearity is to be obtained then phase linearity is also essential, and that both are essential if waveform structure is to be preserved.
Construction of Linear Phase and Non-Linear Phase Systems for Measurement and Subject Comparison

It was said earlier that evidence would be produced showing that measurements related to listening experience. In order to investigate thoroughly the behaviour of both linear and non-linear systems in the listening environment it was necessary to have two systems as near identical as possible, with only one parameter changed.

FIG. 2
I
Stepped baffle used in linear phase' design.

\

I K T i m e delay

2 ;FIG.

FIG. 3 Block diagram of equipment used for measuring phase response of loudspeaker systems.
Acoustic centre

10K 1oK

47K

I
;
FIG. 4
Circuit used to sum output from highllow pass networks.

OUTPUT l f 2

;
8

I

(

LJI\
' \

\

L

FIG. 2

FIG. 4

J

I

FIG. 5
(a) Summed o / p o f 1st o r d e r complementary networks, upper trace ~ / p . l o w e r trace summed o/p: (b) amplitude and phase characteristic 1st o r d e r network.

FIG. 6
(a) Summed o i f r o m 2nd o r d e r com~ plementary networks, upper trace i/p, l o w e r trace summed o/p; (b) amplitude and phase characteristic of 2nd o r d e r network. FIG. 7 (a) 3 r d o r d e r complementary networks, upper trace i/p, l o w e r trace summed olp; (b) amplitude and phase characteristic o f 3 r d order network.

FIG. 8
Amplitude and phase response of linear phase loudspeaker.

10

This project was undertaken after the design of the linear phase model and it was decided to construct a cabinet of identical internal volume to the linear phase design but with a flat, as opposed to stepped baffle plane. Unit selection for the non-linear phase model was made by the widest batch sampling so that tolerances ,of, units were held within k0.25 dB and the transient performance of the two sets of units was identical. Having identical units and an identical cabinet, there were then two options. Firstly, to redesign a crossover system for the non-

linear phase model to give a flat amplitude characteristic, or alternatively, to use the first order networks already designed for the linear phase model, holding tolerances to the tightest limits. The latter course was adopted, as it was felt that any other method would introduce another variable, namely the transfer characteristic of the new crossover and filter network. It should be stated that 'all-pass' networks ahead of the linear phase system were unacceptable as they produce a constant slope phase shift and not the stepped change occurring when time distortion is present.

It was accepted, ,of course, that some departure in measured amplitude response would occur in the non-linear phase system, but as identical drive was applied to identical h i t s this was not counted a relevant variable. The amplitude and phase response of the linear phase model are shown in fig. 8, and the two models illustrated in figs. 9 and 10. Acoustic Environment Not many years ago one heard at audio exhibitions and similar gatherings the statement 'What is the use of taking measurements in anechoic chambers as we listen in normal reverberant surroundings?'. The answer is quite simple, as the measurements we take in anechoic chambers provide much valuable information on how the loudspeakers will sound. The listening experience is essentially a two-part time dependent operation. We receive the direct sounds from the loudspeaker followed by a series of

reflections from the surroundings in which it is situated. It may, therefore, be stated that the free field characteristics are clearly audible in a variety of listening rooms and, equally, the characteristics of the listening room, whatever the loudspeaker used. This fact is also borne out by taking a demonstrably bad loudspeaker and equalising its response in the listening room to the characteristics of a good loudspeaker; the faults in the poor system are still apparent. If comparison is made between figs. 11 and 12 it may be seen that the reproduction of a rectangular pulse is independent of environment, but in the case of the reverberant room, reflections will occur at a discrete time depending on the reverberation characteristic of the room. In fig. 12b the first reflection can easily be seen. As a matter of design necessity the listening distance and height of a linear phase loudspeaker are chosen at the ciesigh stage. Those who argue against the linear phase concept suggest that it is only at this point the advantages gained in waveform performance are apparent in the listening room. This is not the case, as demonstrated by the square-wave performance in figs. 13-16. These oscillograms are of particular interest in that they show clearly the first arrival time component is little affected by room position, but the reverberant field does change considerably. Listening experience whilst these tests were carried out correlated primarily with first arrival signal information. However, if the 'linear phase' loudspeaker is to be of value to the listener it must be shown that current recording

17c

processes reproduce these waveforms from disc and that they are capable of accurate reproduction by the loudspeaker. The accuracy with which reproduction of the complex input has been reproduced (fig. 17) by the 'linear phase' loudspeaker is remarkable, and equally it can be seen that in the case of the nonlinear phase design the first 5 ms of the information has been totally changed in character, and the input waveform destroyed. Bearing in mind the evidence of how first arrival signals show the preservation of wavefronts in reverberant environments, there would seem to be a conclusive case for stating t.kT&fYuoless phase linearity is preserved, the loudspeaker has little chance of accurately transmitting transient waveforms to the Ijstener.
Subjective Listening Tests

The following m a t ~ r i a l was selected for the listenbg comparison: (1) 250 Hz square-wave. (2) Sheffield Disc S10. (3) Musical Box, in anechoic chamber approximately 15 cm from microphone. The listening tests were carried out by random selection of the listening panel, individually, over a period of 21 days. The listener operated the changeover when switch at will and ~ndicated

'

he wanted to move on to the next item in the listening test. The question asked the listener was deliberately kept simple, comprising a two part query. Firstly, can you hear any audible difference between loudspeaker A and loudspeaker B?; secondly, if you can hear a difference would you care to comment on it? Dealing with the results of these tests in the order presented, test signal one was despatched in a matter of seconds .and a considerable difference in what was expressed as 'timbre' or 'tone' was unanimously heard by the unanimity was listeners. E q ~ ~ a l expressed with the second and third test items, although the time taken to make a decision varied widely from listener to listener. Insofar as item two was concerned differences were expressed between slight, almost inaudible, to ' a totally different loudspeaker', and such terms as different attack, greater freedom from coloration, cymbals sounding totally more natural, were three expressions used. With the 'mi~sical box' experiment the terms of expressions were different, and there was fairly general agreement in expressing the difference as better definition, and as if a mask had been removed from the linear phase system as opposed to the non-linear phase model. Dealing with the listeners reactions to items two and three,

it was interesting to note that those engineers who had previous experience in listening comparatively to linear and non-linear phase systems quickly and more definitely identified 'phase distortion'. t his is not, perhaps, surprising as it confornis fairly weil to most listening experience where perception appears to be directly based on experience. The owner is much more aware of a 'rattle' in his car than the passenger who rides in it for the first time.
Conclusions

I

listening room, microphone a t 3 m distance from t h e loudspeaker. off axis horizontally by I m, R H side. FIG. 14 As fig 13 but microphone displaced I m to ler;.

I:$g,':3 displaced 25
but microphone vertically by cm from 'on Axis' ~; 7 . : i Microphone distance reduced t o 1.5 m ~_n_axi:-with design point.

1
I

rib. 1 1

(a, complex wave i i p ; (b) Linear Phase alp: (c) normal alp.

From the work carried out in the preparation of this paper and experience gained in the development of the Linear Phase system, the following conclusions were made. (1) If a loudspeaker is presented with minimi~mphase signals, phase distortion is clearly , audible. (2) Differences are clearly audible when presenting the loudspeaker with 'conventional' material, but the subjective effect varies widely and appears to be more noticeable on material which is recorded with less phase distortion-i.e. PCM or direct cut disc material. (3) In evaluating the transient behaviour of loudspeaker systems by modern instrumentation the measuring problenls

are enorn~ouslysimplified if the system is of minimum phase character. I t is easier to count the horses than the legs and divide by four! (4) Even the anti-phase lobby cannot argue phase linearity is wrong, and if at the lowest level it enables designers to produce better loudspeakers which more accurately recreate the original sound, it is worthwhile.
Acknowledgements

Both in the preparation of this paper and more particularly in the development of the linear phase system-B & W DM6acknowledgement is especially due to the team who were responsible for its design: Dennis Ward, Malcolm R. King B.Sc. (Eng), M.Sc., Stephen Roe B.Sc.
References 1 . M . R . Schroeder: ' M o d e l s of Hearing', Proceedings of the I.E.E.E. September 1975. 2. V i l l y Hansen and Erik R ~ l r b a e k M a d s e n : 'Threshold of Phase Detection by Hearing' presented at the 44th Audio Engineering Society Convention, February 20-22,1973, Rotterdam.

Reprinted from Hi-Fi News and Record R e v i e w . April 1976 b y permission.