Report on talk by Mr. Dan Lavry of Lavry Engineering

for the Audio Engineering Society – India Section on 12th October 2007

'An Evening with Dan Lavry'

A “Design Guru” of our times,  Dan Lavry has been designing electronics equipment for last 37 years.  Mr. Lavry is responsible for many innovative designs including AD/DA converters, Sample Rate Converters, Mic Pres, etc. He is a pioneer in the study of jitter, a concept he brought to the audio industry from his medical equipment design days.

Dan’s designs (widely marketed by Apogee Electronics from about 1990 thru 1998)include AD and DA converters, his old dither called the UV, enabled Apogee to become a converter company not just a filter company.Dan’s own company, Lavry Engineering, is where he manufactures products containing his newest cutting edge circuits and patented designs. Those products are used worldwide by many of the top mastering and recording studios. 

The above short brief about Mr. Dan Lavry in the invitation was a major attraction.
AES India section had the above program on 12th October, 2007 – Friday, at Anupama Audio Visuals studio in Andheri East, Mumbai.
Twenty five members and five guests attended the presentation.
Mr. Lavry and Mrs. Priscilla Lavry were gracious enough to make a stopover at Mumbai, on a request from AES India section.
Mr. Lavry started off with analog and its problems and said that digital solves those problems but has its own set of problems.
To elaborate on this, he explained briefly the process of sampling and quantizing. He said that in the initial years recording engineers complained of a ‘harsh and edgy’ sound of the commercially released digital recorders.
Then it was thought, and rightfully so, that if we have more samples per second we shall have better sound. And more and more samples is better and better, may be. But then that is not correct.
In fact we don’t need to go higher than the Nyquist limit.
Then he went on to support his statement with a very elaborate explanation. He showed the quantization error, in time domain, at Nyquist limit, and then showed the same error in frequency domain. And it was really clear that the error is actually in very high frequencies. And if these error components are removed by high class filters there is really no need for higher sampling rates!!
He talked about aliasing and supported it with excellent visuals. He explained the effect of ‘weak’ filters and showed how like a mirror the images of anything above Nyquist are reflected back in the audio band. And all this makes a very annoying, unmusical sound.
Designing of steep anti-aliasing filters is expensive,  and suffers from all the ailments associated with analogue electronics.
Then in 1990 came the concept of oversampling. Oversampling is a technique whereby
some of this filtering may be done (relatively cheaply and easily) in the digital domain. By sampling at a high rate (for example 4 times 44.1kHz, or 176.4kHz) the analogue
 filter can have a much lower slope since its transition band is now  20kHz to 88kHz (ie half of 176kHz). The samples are then passed   through a digital filter with a sharp cutoff at 20kHz, after which three of every four are discarded, resulting in the sample stream at
44.1kHz that we require

Oversampling and upsampling are techniques that enables simplifying the analog filtering (in both AD and DA). The above is an example of 4 times oversampling. Much of the gear today uses 64-1024 times oversampling, so the transition band has a much lower slope thus the analog filter is less complicated.

When asked that what is the most optimum sampling frequency? Mr. Lavry replied that it would be 60-65Khz.

Then He talked about the mechanical transducers, microphones, speaker which seldom have real response far exceeding 20 kHz. In fact, some of the most highly regarded large diaphragm condenser microphones often used in very high quality recordings seldom exceed 18 kHz bandwidth. So while it may be possible to send very high frequency signals through both analog and digital reproduction chains, there are, in both technologies, fundamental and insurmountable limits to the actual reproducible bandwidth.
He maintained that ‘we need double blind ABX, and doing so shows that very few people can hear in the low 20KHz. Using 88.2-96KHz yields over 40KHz audio bandwidth - way more the we will ever need’.

Moreover digital systems using properly implemented oversampling techniques have far less severe phase and frequency response errors within the audible band

In DAW,s the DSP equalizers and compressors make use of fast processors and due to this low frequencies start to suffer.

Then he talked about ‘Phase’. In that he explained, with very comprehensive visual aid, how the resultant waveform changes with additions of different harmonics and also when their time relationship is changed.
He made a distinction between analog and digital circuits. He said that we need more bandwidth for the ANALOG circuits. Depending on the circuitry, we may need 40KHz ANALOG bandwidth or even much higher. One can work hard at lowering that limit, but with analog, we need some margin. Much of digital circuitry can be done with linear phase without bandwidth restrictions. 

In DAWs when we apply any nonlinear process such as a compressor, and that too having a fast attack, or a process like ‘tube emulation’ then higher harmonics distortion occurs. The distortions at frequencies above Nyquist become alias distortions. Those distortions are difficult to avoid
On the other hand  analog higher frequency distortions beyond hearing are non audible and if so needed they can be filtered.

He then discussed at length dynamic range and bits.
The number of real bits out of a converter is no more then 21 bits. Yet, for DAW work, one may need a lot more bits for the “intermediary processing”. The final outcome of the processed music is then reduced to a final format (typically 16-24 bits). Given that the dynamic range of each original tracks was limited to 21 real bits each, the final outcome is never more then 21 real bits. 

Then he explained the terms ‘Dither’ and Noise Shaping’. And concluded that always start and continue with higher bit rate and down convert your sound at the very last stage with dither and appropriate noise shaping.

Then he came to the topic of ‘Jitter’. He said that he can speak about it for hours, which can be a separate lecture. But then explained briefly and accurately the basics of the problem of jitter. He said that it is important in all conversions but particularly so in A/D conversion. Because once it gets in there then the error is there forever. He explained how jitter is higher for large amplitude high frequency signals. Such signals have the highest slew rate (high slope next to the zero crossing).

Different D/A converters yield different sound quality because of the jitter problem. Different optical media such as CD, CDR may give different quality because of this.
Ofcourse jitter is only one of the D/A issues. There are also other causes for sonic differences.

Dan then discussed the issue of ‘Latency’
After a brief discussion he concluded that unless you are having a live session monitoring latency is not a problem. But equal latency in all channels is desirable than the low latency. With shared signals two channels should exhibit equal latency – this will have better sound when we mix the channels. He played a few recorded samples of music from Argentina.

While concluding he said that there is no such thing called ‘digital sound’. If we have high quality converters then actually there is no difference in a good analog sound and its digital version.

It was a very long session and lasted for more than three hours. It was having a fair number of questions from the members, and Dan was more than happy to satisfy them all.