DVD-Audio Meridian Lossless Packing: The Great Filter Debate

James Guthrie addresses the audience at the ‘Dark Side of the Moon’ launch party in New York.
DVD-Audio cannot be considered ‘high resolution’ – James Guthrie’s speech at the ‘Dark Side of the Moon’ press event in New York impressed upon attendees, including High Fidelity Review’s own Brett Rudolph.

Addressing the contentious subject of whether to release Pink Floyd’s classic album on DVD-Audio or SACD, James cited a number of reasons why he felt Super Audio CD was preferred.

“My principle concern about DVD-A is in the confusion that surrounds it. I speak daily to consumers and industry professionals alike, and I can assure you that the lack of understanding is very real,”

James told High Fidelity Review this past Thursday, reiterating part of his New York oration.

This is certainly a valid issue, but it was his later remarks about the Meridian Lossless Packing process (the compression scheme used by DVD-Audio) that have subsequently had the most impact upon post-event discussions.

“The second issue I have with DVD-A relates to the MLP lossless encoding that apparently has not yet been optimised,” James told the New York audience. “Quick-fix high-frequency filters on the final product, also defeats the concept of a high-resolution format.”

David Kawakami, the SACD Project’s North America Director, was also at the event and as representative of the SACD cause, I asked if he could add a little more detail to James’ comments. “

I believe what James was referring to in his remarks at the ‘Dark Side of the Moon’ launch was the use of high frequency filters to improve the packing efficiency of MLP,”

he told me.

“This practice to reduce audio bandwidth to gain program length is described in various published documents.”

The PDF documents cited by David are ‘The MLP Lossless Compression System’ by M.A. Gerzon, P.G. Craven, J.R. Stuart, M.J. Law and R.J. Wilson, and the ‘Meridian MLP Encoder User Guide’, part of the MLP encoder package.

Having read both documents, it seemed apparent to me that low-pass filtering was not an arbitrary choice made by the MLP encoder, but a pre-encoding choice made by the mastering engineer to enable greater disc capacity. Similarly, an engineer could reduce the sample rate or decrease the bit depth.

David disagrees “…the use of filtering to reduce audio bandwidth is a practice acknowledged and even recommended by MLP itself.”

“If there is no high-frequency filtering in DVD-A, then how do you account for numerous titles in the marketplace that have ‘brick-wall’ filters at 20kHz?” James Guthrie added, although he declined to offer any specific examples. “When I have asked this question, I have been told that these titles had problems with supersonic noise in their files, and that filtering was a quick solution. Indeed, my understanding is that MLP recommends either high frequency filtering, or reduction of word length in order to fit longer program material onto a disc. As MLP can’t compress random noise (which is generated by all Delta-Sigma ADCs), then it stands to reason that this will be an issue with all musical content that has a long run time.”

Thanks to both James and David, there was no doubting their position and current understanding of the DVD-Audio authoring process, but David Kawakami added: “I don’t have a lot of first-hand experience with MLP because I haven’t made any DVD-Audio titles. I would suggest you contact some mastering engineers who have experience with MLP encoding.”

An excellent suggestion, one that could also enable us to learn more about the intricacies of the encoding process. Who better to ask for their comments than those who produce high-resolution titles on a daily basis, so we turned to John Kellogg, General Manager of Multi-channel Audio and Music at Dolby Laboratories who produced DVD-Audio titles for Deep Purple, Foreigner and Emerson, Lake and Palmer, Craig Anderson, DVD Development engineer at WEA Studios who has mastered dozens of DVD-Audio titles for the Warner Bros. group of labels, Chris Haynes, a Chief Engineer at 5.1 Entertainment, a company that has almost two hundred DVD-Audio titles under their belt, Mark Waldrep, Ph.D. of AIX records, who has produced over fifty DVD-Audio titles for his own label and the likes of Denon Classics, including the award-winning ‘Nitty Gritty Surround’, and the industry’s mastering ‘guru’, Bob Ludwig of Gateway Mastering and DVD, who has been involved with countless DVD-Audio and SACD titles.

For an explanation of the technical intricacies of Meridian Lossless Packing, High Fidelity Review also spoke to the co-author of the paper referenced, Robert Stuart of Meridian Audio.

John Kellogg: “James Guthrie is an immensely talented producer and engineer who has produced, in my opinion, some of the best recorded music available, in stereo or multi-channel. But James has no direct experience with MLP encoding and, to the best of my knowledge, has yet to create a track for DVD-Audio release.

“It is unfortunate that Mr. Kawakami has chosen to disseminate misinformation to Mr. Guthrie and others regarding MLP – something he also has little applicable or practical knowledge of and I suspect no direct experience in using. It is absurd for him to state or imply that DVD-Audio is not a high-resolution format or that MLP somehow prevents or compromises high-resolution audio.”

Dolby Laboratories are responsible for licensing Meridian Lossless Packing to the industry, and as such their technical and development departments are intimately familiar with the software. “MLP does not require filtering or reduction of word length and neither are recommended by Dolby. If a file is corrupted or troubled (where it is flagged as such by an MLP encoder verification process), then the solution is to go back and create a clean file,” John explained.

Craig Anderson: “Regarding the supposed filtering in DVD-A, it seems that Mr. Guthrie has been misinformed about the MLP process. Of the fifty-plus DVD-A titles bearing my name, none has been filtered in any way between the mastering stage and the MLP procedure. It’s simply not necessary, nor is it recommended. MLP has proven to be a very robust and reliable means of delivering a superior quality audio. It has so far been, as its name implies, lossless,” he explained. “It seems David is confusing ‘recommended low-pass filtering’ on MLP and ‘required low-pass filtering’ of DSD.”

The specifications for Super Audio CD require a 50kHz low-pass filter in all playback devices, which removes content above 50kHz to avoid undesirable oscillations brought about by the high-frequency noise from Direct Stream Digital’s aggressive 42dB per octave noise-shaping filters.

Chris Haynes: “It is dither in a 96kHz environment that means death to MLP encoding. The MLP process removes correlated information; dither is noise and noise cannot be correlated. The presence of dither means no MLP without removing the dither through filtering. If you are going to MLP [encode] a PCM file, you cannot use dither at a 96kHz sampling rate.” Haynes, who is finishing up the DVD-Audio production of Queensrÿche front man Geoff Tate’s self titled solo album, wondered whether Guthrie “…knowingly or unknowingly introduced dither after his audio became 96kHz PCM? Some equipment has default dither that must be operator defeated.”

However, there are ways around this problem if it occurs, that minimise the use of filters but still allow the 9.6Mb/s data throughput of DVD-Audio, Haynes explained to High Fidelity Review reporter Sanjay Durani when he visited their West Los Angeles facility last Thursday. One of the tools used by 5.1 Entertainment is the SurCode MLP encoder from Minnetonka Audio Software (see our recent news story), which allows the operator to take some of the bit-pool from channels where the loss will be least noticed (the LFE for example) and use it for channels where it might be more needed (the front left and right pair). While Guthrie’s remarks might be accurate from one perspective, the problem isn’t seen as a real-world impediment to releasing high quality, high-resolution music on DVD-Audio; certainly not from people who work day-in and day-out with PCM, MLP and DVD-Audio at 5.1 Entertainment. Jeff Dean, who was recently named President of Silverline Records, feels that Guthrie may have been “…theorising potential problems rather than talking from real working experience.”

Bob Ludwig: “All great music recordings are of necessity the result of intelligently made compromises,” he told me, “…but having said that, Meridian Lossless Packing provides bit-for-bit zero-loss data reduction so the packed and then un-packed signal is a bit-for-bit clone of the original signal. So, we have a system that can work perfectly.

“Just as in vinyl disk cutting days, if a producer wishes to have a vinyl disk or DVD-Audio disc with a longer than usual side length, compromises need to be made to stretch the physics of the system. With vinyl disks, usually bass reduction filters were employed, or sometimes a mono-ing of the low bass to prevent vertical excursions.”

So can engineers use similar tactics today? “When it comes time to stretch the laws of physics to accommodate an unusual DVD-Audio title the following is possible with MLP encoding software: We own two different Meridian approved MLP encoders, one from Meridian themselves and another from Minnetonka software. The only option built into these programs is to reduce bit depth of an individual channel. How audible an LFE woofer channel would sound being reduced from the 24-bit original to a new output of 20-bits I would personally defy anyone on the planet to reliably detect! More severe reductions would be to reduce the bit depth of the rear channels to 20-bit etc. Any filtering of high frequencies, if done, was always done with third party filters, never was it part of the commercially issued MLP encoders. How often does the need come up to change these settings and choose to not have a bit-for-bit clone? Not often! The vast majority of titles needing MLP encodes sail through perfectly the first time, thank you!”

Which brought us to the supposed use of a high frequency filter in the MLP encode chain. Bob drew my attention to the following paragraph from Meridian’s AES paper: “A typical 96kHz 24-bit six-channel program would encode to an average of 7.2Mb/s, reducing the audio bandwidth with simple filtering from 48kHz to 24kHz will generally reduce the rate to below 5Mb/s.” “This is physics,” Bob added.

And what of real-world use? “As a practical matter there is no problem with MLP encoding. I think the odd title here and there that may have had problems encoding are, especially these days, an extremely rare occurrence. We routinely MLP encode DVD-Audio reference discs of material originally made on DSD converters (there are no SACD ‘DVD-R’ recorders commercially available) which have large amounts of noise shaped energy put into the ultrasonic band (60kHz). Again, with absolutely zero loss.” As has already been mentioned, DSD is inherently full of high-frequency noise, so to learn that MLP can transparently encode what must present a near ‘worst case’ scenario, adds another dimension to the issue.

“Please don’t forget,” Bob Ludwig added, “…in the real world, time-is-money deadlines loom over everyone’s head. A MLP encode engineer might intentionally, on their own, choose to slightly compromise the integrity of the music of a difficult encode in order to meet a deadline, rather than taking the time necessary to go the two or three attempts in order to do it with zero loss. From our own DVD-Audio authoring experience where the front office asks the encode engineer why a MLP encode was not done ‘on time’, the vast majority of MLP encodes fail due to the engineer setting the stereo down-mix coefficients too aggressively, which therefore causes a digital ‘over-level’ when the six channels are combined than from any other cause.”

Mark Waldrep, Ph.D.
thanked High Fidelity Review for allowing him to shed a little light on what he described as a difficult technical and political situation. “I read with astonishment the comments made by James Guthrie about MLP and the DVD-Audio format… somehow he’s been seriously misinformed about both the theoretical and practical application of MLP by mastering engineers and encoding specialists.” Mark is clearly both. “I honestly doubt that he’s had any practical experience with Meridian’s brilliant algorithm or personally talked to anyone actively working with these tools.”

Illustrated by the diverse genres available from AIX, Mark explained that between the tracks he prepares for his own label and others, that he has encountered many different audio qualities and issues. “In no case have I ever used either high-frequency filtering or word length reduction in the preparation of a DVD-Audio project. It’s simply not required to make a proper MLP encoded sound file. In fact, the use of extremely high sample rates (96kHz and 192kHz) obviates the need for harsh anti-aliasing filters as well… the thought of resorting to a ‘brick-wall’ filter or other bit modifications to fit longer program material on a disc has never entered my mind. I route high-resolution digital audio directly from a Sonic Solutions EDL into the ‘One Click’ MLP encoder and initiate the processing without any shortcuts or intervening signal processing. The MLP encoder does its job and that high-resolution audio stream is authored into the final disc image.”

Clearly disturbed by the implication that his DVD-Audio productions may somehow be technically flawed, Mark stressed that he takes great pride in the fact that AIX Records is producing audio of the highest quality. “Why would I or any other engineer working in this area compromise this quality in the launching of a new audio format? DVD-Audio and high-resolution PCM encoding, with or without MLP, are unique in providing me the audio quality and work flow that I’ve come to appreciate and know over twenty-years of audio engineering. When combined with DVD-Video compliant audio streams and multimedia, there’s simply no other format that comes close.

“I have only the greatest respect for James Guthrie and his career,” Mark added. “Unfortunately, for reasons that we may never know, he has made some rather uninformed comments about a format and an encoding methodology (MLP) that have been widely circulated and taken as fact by many equally uninformed individuals.”

It seems therefore, that if the experiences of those who have produced hundreds of DVD-Audio titles between them are anything to go by, that pre-filtering is a non-issue and that encode filtering is neither recommended nor used by any of the engineers who spoke to High Fidelity Review. The overwhelming impression was that the MLP system could certainly be abused as a result of careless production values, which is the case for all delivery formats, but can and does, to borrow the words of Bob Ludwig, “work perfectly”.

But what about the technicalities of the system and the claims made against it?

Robert Stuart: “James Guthrie has implied that MLP is not optimised and that ‘brick-wall’ filters are required to make a DVD-Audio disc. This is categorically not the case, as any of the labels actively producing DVD-Audio will testify. Whilst I have never spoken to Mr Guthrie, I am told he has never worked with MLP, and so I imagine he has been misinformed,” a view that was a recurring theme amongst all those with whom High Fidelity Review discussed the issue.

“MLP is not only mature and stable, but it was selected for DVD-Audio (and DVD-Audio Recordable) because of its very high performance and strong feature set. It is a true lossless system, which means that the output is a bit-accurate recovery of the master.

“Since we are strong advocates of high-resolution audio, we encourage engineers to produce clean masters and to avoid badly-behaved tools. Beyond that we do not presume to suggest how they do their work. When the master is complete and all parties are happy with it, the job of MLP is to transfer the data losslessly, which it does.

“It is ill-advised for David Kawakami to comment on MLP as he is neither an expert on audio coding, nor on MLP.” This was something David Kawakami was himself, quick to point out. “However, since these points are raised I would like to explain where his understanding is flawed.

“All lossless compression systems, including those that compress 1-bit streams such as DST, are limited to the ‘entropy’ or ‘noise-floor’ in the incoming signal. This is not a problem with any technique, it is a fundamental aspect of information theory. If a signal contains a lot of wideband or high-frequency noise then you end up with a bigger file. More compression would result if this noise were reduced or, much better, not introduced in the first place. In our MLP training information, we point out as a matter of information that the size of a compressed file can be adjusted by using (gentle) low-pass filtering or selection of a word size to suit the project. This is useful background information to a certain type of producer, who may want to free up space on a disc for other assets or simply understand how the process works.

“Since these points apply to all lossless compression methods, filtering can also be used to adjust the playing time of SACD releases. Although we have no experience of making these discs, information theory presents a level playing field.

“However, there is a huge gap between explaining the choices available to mastering engineers and them being encouraged or required. Modern mixing desks contain many kinds of filters, EQ units, reverbs etc., that are used to produce the desired artistic result; that is not our concern and no-one forces their use. What we are doing is explaining the impact of such choices on the project.

“When the final master is ready, it is the job of MLP to deliver it exactly. To guarantee this, the MLP workflow includes a proofing step, which establishes bit-accuracy. Furthermore, the player actually checks that the transfer process from master to output is lossless. This ability, to actually deliver the full master quality, is unique to DVD-Audio.”

Robert then went on to raise a point that we ourselves had been wondering about and had put, without receiving a response, to David Kawakami and is mentioned earlier in this piece: “It is worth pointing out the irony in this filter discussion: due to the very high supersonic noise power in a DSD stream, SACD players all incorporate (usually steep 50kHz) low-pass filters to prevent damage to downstream replay equipment. The presence of these filters automatically renders the replay process lossy, not lossless.

“In summary, MLP is highly stable, dependable, and has been successfully used in the overwhelming majority of the five hundred or so DVD-Audio releases to date. Meridian does not advocate the use of filtering in routine DVD-Audio production or in playback; rather we encourage producers to avoid errors in source files. We are also not aware of filtering being used as part of the authoring process by any of our customers.

“James Guthrie also alluded to the use of brick-wall filters in DVD-Audio production. We have no idea what he is talking about or why it would be needed. In fact, as has been recently demonstrated, linear-phase (also known as ‘brick-wall’) filters are completely unnecessary in audio PCM systems running at high sample rates like 88kHz and above.”

And how good is PCM packed with MLP? “The coding space provided in DVD-Audio with 96kHz and 24bits is rectangular, with a dynamic range of 144dB that pertains over the entire 48kHz audio bandwidth,” Robert commented. “This means that the finest musical nuances can be delivered on the disc without the need to compromise by either chopping off high-frequency information or swamping it in a wash of noise. No other music carrier can approach this. Certainly single bit coding can’t; its range is only 120dB up to 20kHz and less than 60dB up to the player roll-off of 50kHz! This, and jitter sensitivity, are the very reasons the high-end gave up using bitstream conversion ten years ago. So, DVD-Audio’s coding space is huge, and to be able to fit six of these channels onto a disc is a spectacular achievement.”

A number of things are clear from this investigation. Firstly, the old adage of ‘garbage in – garbage out’ still holds true and that there is no substitute for carefully produced and recorded source material. Secondly, none of the producers and engineers we spoke to had found the need to filter source files prior to encode, and under no circumstances are filters “recommended”. In some cases filtering may well be a “quick-fix”, but a “fix” for flawed, sub-standard pre-production work that none of the engineers mentioned here would accept regardless of format, and not the encode/decode chain per se.

The last word, at least for now, belongs to Robert Stuart:

“James Guthrie is a highly respected producer whose work I have enjoyed. I am confident that if he decides to prepare content for DVD-Audio, and if he takes advice from experts, then he will appreciate the flexibility and integrity of MLP and be more than satisfied with the sound of the disc.”