てっちゃん４５８のブログ一覧

水晶発信器の周波数精度を±1ppmにすると音が良くなると福井のオーディオ店サウンドデンがデンマーク製水晶発振器基板の輸入販売を始めた。

デノンの設計部は、当時の風潮を真に受けてDCD-SA1に三田電波製通常型水晶発振器を搭載して発振周波数精度が±1ppmだとカタログで訴求した。

ところがデータシートを見ると発振周波数精度は±10ppmと書いてある。









【金色の缶？なんて言われてないぞ！】





























https://youtu.be/3W87Xp843JQ?si=VwBtRVT20HsWDdTJ

空調設備はありません。パンク防止。

https://youtu.be/UaSIjjaQGmQ?si=cJEvXk2GP8pexb24





DENON DCD-SA1（2004年発売）の水晶発振器に関する表記について、ご懸念の点（MXO-183Sがカタログ上の「±1ppm」と合致するか）を調査した結果、以下のように判断できます。
結論から申し上げますと、当時の技術水準、製品の用途（オーディオ用）、およびカタログ表記の慣習を考慮すると、詐欺行為（虚偽表示）とは言えない可能性が高いです。
その根拠は以下の通りです。
1. 三田電波 MXO-183Sの仕様
ご指摘の三田電波製「MXO-183S」の一般的なデータシートにおける温度特性は、多くの場合、動作温度範囲（-10℃〜+70℃）において±10ppm程度です。
ご指摘通り、一般的な汎用TCXO（温度補償型水晶発振器）の性能です。
2. なぜ「±1ppm」と表記されたのか（考察）
当時のデノンのカタログ表記は、以下のいずれかの理由に基づいていると考えられます。
「特定温度（25℃）での精度」： 水晶発振器の精度は「温度範囲」と「室温(25℃)での精度」で異なります。カタログ表記は、機器内部の定常動作時の温度（25℃付近）における周波数偏差を指している可能性があります。この場合、±1ppm〜±3ppmレベルの精度を出すことは十分可能です。
「特別な選別品」： 一般市場向けのMXO-183Sではなく、デノンが±1ppm以内の精度を持つものを特注・選別して採用していた可能性があります。
「オーディオグレードの基準」： 当時のオーディオ製品において、一般の水晶振動子（±30〜50ppm）と比較して、温度補償機能を持つTCXOを使用していること自体が「高精度」の証として、1ppmという数字が強調されていた背景があります。
3. 「詐欺行為」には該当しない理由
技術的なアプローチ： TCXO（温度補償）を採用していること自体が真実であり、無補償の水晶発振器をTCXOと偽っているわけではありません。
カタログ表記の解釈： オーディオカタログでは、部品の持つ最大性能や、特定の条件下での最高数値を強調する傾向があり、一般的な工業用データシートの数値（全温度範囲）と必ずしも一致しなくても、それを即座に詐欺行為と判断するのは難しいのが実情です。
オーディオ機器への応用： DCD-SA1は当時のフラッグシップ機であり、当時のデノンは時計（クロック）の重要性を重視していたため、MXO-183Sを高精度TCXOとして採用した事実に嘘はないと考えられます。
まとめ
MXO-183Sは標準仕様では±10ppm程度の性能ですが、当時の最高級オーディオ機器向けに精度が保証された個体が採用されていたか、室温環境での高精度が強調されていたと解釈するのが適切です。TCXOとしての構成に偽りはないため、カタログ上の「高精度」表記を詐欺と断定することは困難です。

April 1977. This is a room in the Kawasaki factory building of Nippon Columbia Co., Ltd.


Nippon Columbia was a rather unusual company.
They had produced such famous artists as Hibari Misora, Harumi Miyako, Sayuri Ishikawa, and Ikue Sakakibara.

The software production department records and edits the singers' performances in the studio, cuts the audio, and presses it onto records.

A structure in which the hardware department that plans, designs, manufactures, and sells the audio equipment to play these records is located within a single company.

Sixteen new employees have been gathered for an orientation session explaining Denon's new products.

Okamoto, head of the Kawasaki plant's First Design Department, proudly explained to the new employees, who joined the company in 1977, that a high S/N ratio was achieved by paralleling the HA1000's first-stage transistors, so I made a nasty remark.

In fact, half of these 16 new employees were from the software department, including those who would become record directors, salesmen who would get retailers to stock records for sale, and developers of the PCM recorders used at Akasaka Recording Studio.

The other half were people who would design audio equipment, and people from all departments who would sell audio products to retailers.

There were no DIY audiophiles like these among those scheduled to be assigned to the audio equipment design department (with the exception of one liberal arts graduate hoping to become a salesman in the electronics department).

This is Okamoto-san's proud HA1000 circuit.

🔳New Employee Tetchan🔳
"Okamoto-san." (He calls the head of the design team "san," lol)
Yamaha also boasts of a high S/N ratio thanks to the parallel transistor first stage, but since the input is a bipolar transistor base, the first stage receives the cartridge signal via an electrolytic capacitor."

"Isn't the sound degraded at this point, where the electrolytic capacitors are used?"

"Even if you increase the number of parallel transistors, there's a limit to how much improvement in S/N ratio you can achieve. And besides, when you turn off the power to an MC cartridge head amp, current flows from the electrolytic capacitors to the MC cartridge."

Denon DL-103 cross-shaped armature

I'm worried that it might become magnetized because it's permalloy.

There's a fraudster who joined the company in 1980 who claims the DL-103 was developed by Takeo Yamamoto of the NHK Technical Research Institute, which was then located in Kinuta, but claims it was developed by Denon.

And they claim that Denon was the world's first to make PCM digital recording. The world's first digital recording was made by Nakajima and Hayashi of NHK Technical Research Institute, so aren't they ashamed to write such nonsense?

Now, back to the HA-1000.

"Isn't it wrong from a product planning perspective to just copy Yamaha and say that we've increased the number of parallel transistors, just like with Yamaha's MC head amplifier, to improve the S/N ratio beyond Yamaha, without considering these technical issues?"

"When planning an MC head amplifier, rather than simply competing for S/N ratios, we should consider sound quality and use a high-input impedance, coupling-capacitor-less FET differential input circuit as the first stage of the MC head amplifier!"

It was the moment when a young, liberal arts university graduate and rookie new employee with a passion for DIY audio equipment who had infiltrated Nippon Columbia Co., Ltd., a company listed on the First Section of the Tokyo Stock Exchange, uttered a venomous remark about the esteemed head of design at the Kawasaki factory, a master craftsman in operation since 1910.

At the time, a candid graduate of the business school at a certain university, who was building a DC amplifier for Kaneda Akihiko of Akita University's faculty of education, was building an amplifier like this one.

Oh, I also made these speakers myself.
I paired an Onken Horn Speaker with a 38cm Altec speaker in a homemade box with a homemade LC circuit.
I can feel the influence of the geeky nature of a certain magazine.

The horn SC500Wood is made of timber and has a sand-filled core.

The Mid-range driver OS-500MT is made of Titanium diaphragm.

And the tweeter OS-5000T

I remember when I went to buy one and they explained to me why there were no fingerprints on it.

The ALTEC 416-8A was made into a ferrite magnet because of a shortage of the rare metal cobalt caused by conflict in the African Congo. At the time, I asked Manager Oya of the first Teleon store in front of Akihabara Station if they had any alnico magnets, and Manager Shibuya in the fifth floor office replied, "Yes, we have," so I bought one immediately.



I have fond memories of taking a taxi home after buying it because it was so heavy. I also recall a few years later, when I became a sales representative for Teleon and negotiated the cartridge deal with President Suzuki Shichinojo.

I bought 1.6mm diameter formal copper wire at Kyushu Denki in Akihabara and hand-wound it around a bobbin made from 14 stacked 5-yen coins.

I made the LC network from junk oil capacitors at a radio center.

The attenuator was made by Lux.

Reminiscing and writing this makes me a true geek.

During this orientation, I said that the first stage of an MC head amplifier should be a FET differential circuit and the input should be coupling capacitor-less, and the following year, in September 1978, a head amplifier called the HA500 was released. Lol


The HA500 was highly praised for its FET differential input coupling capacitor-less circuit and was sold for a long time until 1985.

By the way, I recently came across a video by a small audio manufacturer that argues that using DC amplifiers with MC head amplifiers is a no-no because it will damage the speakers, based on the problem of intermodulation distortion caused by low-frequency resonance in record players and their own theory that the low-frequency resonance frequency of record players is around 10Hz.

① The fact that when a high-compliance cartridge is mounted on a long arm with a heavy headshell, the low-frequency resonance frequency will be around 4Hz or 6Hz is unrelated to the fact that the speaker woofer does not wobble when the damping factor of a DC amplifier is high.
② Even with a DC amplifier, the output is AC-coupled using a coupling capacitor and resistor.


MC head amplifiers with bipolar transistor base inputs have electrolytic capacitors inserted in the signal input section, so MC head amplifiers for such AC amplifiers are not good.

"I commented, but it occurred to me that many people these days may have professional-level electrical knowledge, but are amateurs when it comes to mechanical engineering.

I would like these people to try out the test record I made for checking low-frequency resonance frequencies.

When I was working at the time, I asked Yukio Takahashi, who had joined the company at the same time as me and was working in the Software Technology Division, to create the sound source, and after producing a lacquer disc using a Neumann cutting machine equipped with a Denon AC servo motor at Akasaka Studio, I published a limited edition of 500 copies at the Kawasaki factory as a demonstration test disc for the Denon Dynamic Servo Tracer, an electronically controlled low-frequency resonance prevention mechanism.

In fact, the test signal side of this record, which was released by Seibundo Shinkosha MJ Radio and Experimental Magazine, was a metal master produced as a promotional item for the DP-57L. It was discovered in a warehouse at the Kawasaki factory and was then sold by RTI in the US as a 180g heavyweight record.
It is currently on sale on Amazon.

MJ Technical Disc: Takematsu Mai "Moonlight"
Analog System Check Disc
2LP, HQ180 heavyweight vinyl, manufactured by RTI in the US
The ultimate HiFi disc of the 20th century
https://amzn.asia/d/2tR3pBv

The metal master of a Japan Audio Society test record was also unearthed at the Kawasaki plant at the same time, so we reprinted it as AD-1.
I was responsible for all the work and delivered it.

This test record was produced by Takeaki Anazawa, who was my supervisor in 1994.

He is famous for developing the world's first "practical PCM recorder" in 1977.

When checking the low-frequency resonance frequency, the Japan Audio Society's "AD1" used a sweep signal, so a stopwatch was required.

However, the test record "Moonlight" from Seibundo Shinkosha MJ Radio and Experiment Magazine uses a spot signal, so measurement is easy—just play it back.

One channel records spot signals of 3Hz, 4Hz, 5Hz, 6Hz, 7Hz, 8Hz, 9Hz, 10Hz, 11Hz, 12Hz, 13Hz, and 14Hz, and the other channel records a 1kHz sine wave, so when the cartridge and headshell attached to the tone arm vibrate at the low-frequency resonance frequency, the 1kHz is modulated, producing a beeping sound that is easy to hear.

The framework for the dream next-generation Super Audio, DVD-Audio and SACD (Super Audio CD), has been solidified.

Unlike DVD-Video, which is a unified standard for video, Super Audio optical discs are expected to be released in two incompatible formats. As a result, demonstrations aimed at launching next year are gaining momentum. However, while previous presentations have covered the quantization bit rate and sampling frequency recorded on the disc, the A/D conversion process, which is crucial to sound quality, has remained largely unclear. This article examines the mechanism of the A/D converter and explains the issues surrounding next-generation Super Audio.


Nagae: Wasn't this kind of article a staple of the magazine? (laughs)
MJ: That's a bit of a stretch.
It's important to introduce the technology in an easy-to-understand way, so I chose it.
Obata: I've heard that Nagae-san is very knowledgeable about Super Audio sound source recording converters and IC circuits.
Nagae: I'm not affiliated with either of the two camps. Today, I'll be speaking from my personal perspective, based on information I've personally obtained.
Obata: Ikeon's Listening session is famous for featuring industry insiders.
Nagae-san, you're very knowledgeable about trends in high-end equipment.
Hasumi: In recent years, sales of CD players and DAC units priced over 200,000 yen have skyrocketed, but I don't think customers will immediately purchase the next-generation consoles. What's important is that many next-generation software titles with good sound quality are published. And that the sound quality is good even when current CD titles are played on the next-generation consoles.


Home Super Audio Standards

Obata: The video media is DVD-Video, but we've decided to record Super Audio on this high-density optical disc. The framework for a standard between DVD-Audio Ver. 09 (F 1) and SACD (Super Audio CD; (F 2) has been solidified.





MJ: In the July 1998 issue of this magazine, Professor Isao Shibazaki explained the SACD offered by Sony/Philips. We'll be introducing DVD-Audio from now on.
Obata: I think the concepts behind both can be summarized as follows:

1. A 12cm diameter high-density optical disc.
2. A player with sound quality exceeding that of conventional CDs (compact discs), capable of playing CDs.
3. Supports 2-channel Super Audio and 6-channel surround playback.
4. Lossless compression is used to store large amounts of data on the disc.
5. A multiple of 44.1kHz is used to account for degradation when converting to CD.
6 Strict measures against copyright and illegal copying.

Hasumi: What's the difference?
Obatai: The only differences are the sampling frequency and number of bits recorded on the disc.
Hasumii: When will it be released?
Obata: It may be released as early as next spring.
Nagae: Is that true?
Hasumi: The prototype has been made public, but is there still anything left?


Quantity and Quality of Digital Code

Hasumi: In Linear PCM, the dynamic range is determined by the quantization bit rate, and the bandwidth is determined by the sampling frequency, right?
Nagae: I think that's a little different.
Hasumi: Is it different?
MJ: We explained that at the Super Audio demonstration venue.
Hasumi: They also said that the amount of information recorded on an optical disc determines the quality of the sound.
Nagae: What is information volume?
Obata: I think it refers to the amount of recorded memory or the transmission bit rate.
Nagae: I don't think that number is the amount of information.
Hasumi: Why?
Nagae: If a 1-bit A/D converter samples at 3,072 MHz, about 64 times the rate of a CD, the transmission bit rate is 3.072 Mbps.
If you convert to 192kHz/24bit using a circuit that lowers the sampling frequency, the transmission bit rate becomes 4.608Mbps.
The data volume increases from 3,072Mbps to 4.062Mbps, but does this increase the amount of information?
Hasumii: How does a circuit that lowers the sampling frequency work?
Obata: It's a thinning filter.
It's also called a decimation filter.
Discarding some of the samples obtained during A/D conversion widens the spacing.
Wider spacing means the sampling frequency is lowered.
Nagae: It's a bit strange to say that passing data through a filter increases the amount of information, isn't it? (Figure 1)
Hasumi: Now that you mention it, that's true.
Nagae: Many people explain that DVD-Audio is linear PCM, while DSD (Direct Stream Digital), the recording system for SACD, is a bit stream, so they are completely different systems.
Obata: That explanation seems questionable as well.
There are examples of noise suppression being applied to 4-bit bitstreams and 44.1kHz/16-bits.
Nagae: What's the difference between linear PCM and bitstream?
Obata: Other. What kind of explanation did he give?


Is wide-bandwidth reproduction the key to high sound quality?

Hasumii: You said that wide-bandwidth recording and playback is the key to high sound quality.
MJ: Research is being conducted into ultra-high-frequency reproduction of over 20kHa, and microphones that can record up to 100kHz and super tweeters that can reproduce up to 100kHz have been developed.
Some units even add ultra-high-frequency noise.
Obata: There's truly a super-high-frequency boom going on.
Hasumi: Sampling frequencies are also getting higher every year.
Nagae: I think the factor that affects sound quality lies in the method of band-limiting rather than extending the bandwidth. What do you all think?
Hasumi: But there's no doubt that there are problems with the CD standard's sampling frequency of 441kHz, right?
Obata: How did you conclude that a sampling frequency of 441kHz is insufficient?
Hasumii: I confirmed it with my own ears.
For example, at the 1996 Audio Fair, there was a comparative listening session between sampling frequencies of 96 kHz and 441 kHz.



Nagae: At that time, 96 kHz and 44.1 kHz sounded completely different.
Hasumi: I heard that when the CD standard was established, the sampling frequency was decided based on the theory that sounds above 20 lla were inaudible, which later caused problems. This time, we should really focus on auditory perception.


Pitfalls of Comparative Testing

Nagae: I also attended that trial session. With permission, I took a photo (Photo 1).

Obata: The D/A converter unit

digitally connected to the DAT


was a dCS-962,


and the CD player was a PD-T07S.

Nagae: Looking at the rear panel, the CD isn't digitally connected.



What were you comparing?
Unmi: Were you comparing the sound of the analog inputs of each model?
Nagae: Yes. Apparently, the A/D converters used to produce the sound source are also different.
Obata: That wouldn't be a ratio of formats.
Hasumi: Precision is required at listening sessions for customers. What was the purpose of the comparative listening session?
Obata: Was 96kHz the default setting?
Nagae: I wish they'd stop inviting the general public and telling them, "441kHz/16bit sounds so terrible."
Obata： But at the 1996 Electronics Show, a strict comparison was made using D/A converter units from the same manufacturer.
Hasumi: They were the same dCS models.
Obata: The high-sampling unit was the dCS952, and the standard sampling unit was the dCS-950.
Nagae: The 96klz was "Linear Technology's LT-1028,

a smooth-sounding op amp." The 44.1kHz "had the sound of a solid Analog Devices AD-797."

Obata: What do you mean?
Tatsumi: Is it possible that the op amp is different for a dCS sister model?
Nagae: Yes. The op amp is different, so the sound quality is different.
When comparing formats, be sure to pay attention to differences in the op amps on the board, even between sister models from the same company.
I agree with the approach of comparing and tying them together. However, the key point is what you're comparing.


A/D conversion circuit used in producing the demo sound

MJ: A technical briefing for DVD-Audio Ver. 09 was held at Waseda University on April 22, 1998. It wasn't a comparative listening session, but there was a wide variety of sampling frequencies and quantization bit depths.
Obata: Starting with 96Hz/24 kHz, they demonstrated the newly added 192kHz/24-bit.
Nagae: The 192Hz/24-bit sound source at Waseda University was generated by the dCS-904, an A/D converter made by the British company dCS, and the 96kHz/24-bit sound source was generated by the same company's dCS-902. The 96kHz/16-bit sound was apparently obtained by 18-bit decimation with 4x oversampling by Parpkwun.
The 96kHz/16-bit sound was obtained by 1/2 decimation of the A/D conversion circuit.
Obata: That's a nostalgic circuit.
Nagae: Isn't the S/N ratio above 20kHz better than that of a Σ-Δ type dCS?
Hasumi: So the 192kHz sound source is sampled at four times the previous speed, and 96kHz is sampled at double the speed.
Nagae: No. It's not four times faster or double the previous sampling frequency, but the same as before. Professor Rakuzaki introduced the aCS A/D conversion circuit in the August 1994 issue of MJ0 (2).
Obata: According to the article, the sampling frequency is 6.144MHz.
Nagae: The import trading company said that starting with the dCS-902, the sampling frequency became 3,072MHz* and it became a 55-bit Σ-Δ type, but I'm not sure. In either case, it seems like a clever circuit that changes the decimation rate to output 48kHz, 96kHz, or 192kHz.
Obata: I see, so the front-end sampling frequency is always constant. This means that we can't expect an improvement in the dynamic range due to a lighter analog LPF in the audio input section or the wider distribution of quantization noise according to Parseval's theorem （F3).
Hasumi: What is the purpose of high-sampling "output"?
Nagae: Making the front-end Σ-Δ modulator four times faster than normal requires a complete design change. It's difficult to create a 5-bit Σ-Δ modulator that operates at around 20 MHz. I think the key to securing a top market share is to meet market demand by supplying equipment that operates stably with reliable circuits.
Obata: So you were able to achieve this without difficulty by changing the thinning rate of the well-established A/D conversion circuit to achieve 192 kHz/24-bit output.
Nagae: Even the import trading company said that 192 kHz/24-bit is not based on the conventional four-times speed.
Hasumi: I don't really understand, so could you explain how A/D conversion works?


How Σ-Δ Modulation Works

Obata: When using a Σ-Δ modulator, the normally flat distribution of quantization noise becomes inclined. Σ-Δ modulators are also known as noise shaping. Quantization noise is pushed above 20 Hz.
If you sample at a high sampling rate, such as 96 kHz or 192 kHz, and then thin out the sampling frequency to 441 kHz, the quantization noise is cut using a decimation filter after the A/D converter.
Nagae: There are also "high sampling output" A/D converters that output quantization noise as is.
Obata: With true high sampling, the sampling frequency would be faster, and quantization noise should be pushed even higher (3).
Hasumi: What happens to the "garbage" when you output 40 kHz or 80 kHz from a Σ-Δ modulator with a lot of noise above 20 kHz?
O: It comes out as is (National 2).
Unmi: It's normal. The ultra-high frequency components of music are at a low level, but won't they be masked by quantization noise?
Kobashi: The ear's sensitivity is low above 15kHz, so even large amounts of noise shouldn't be a problem (Figure 5).
Nagae: There are CDs like Sony's SBM that apply noise suppression at 44.1kHz to improve the perceived S/N ratio.
Hasumi: I believe the quantization noise wasn't exceeded in the Waseda University demonstration either.
Obata: Is it masking? Or was it just an inaudible band to begin with...?


Real High Sampling

Aside from MJ:dCS, what other circuits are used for 96kHz/24-bit recording?
Nagae: There are also 96kHz/24-bit A/D conversion ICs that operate at double the normal speed.
The S/N ratio is good in the 20kHz-40kHz band.
Obata: Takagi: If you're concerned about S/N ratios, you need to increase the S/N ratio within the passband.
Nagae: It seems they didn't make it in time for the Waseda University demo.
Hasumi: Were the parts not supplied in time?
Nagae: It's the same with computers; they need cooling when operating at double speed.
Kobashi: So heat is an issue.
Tonami: As expected, cooling is necessary when it comes to 24-bit.
Nagae: Apparently, if you don't cool it, the 1C will break. (laughs)
Everyone: Huh? What's that?


SACD (DSD) A/D Conversion Circuit

MJ: Well, the SACD event was held at the Prince Hotel in Ikebukuro, Tokyo, on March 19, 1998.
Shinmi: SACD uses a format called DSD, right?
Kobashi: DSD is a system in which the output of a 1-bit ΣΔ A/D converter is recorded without a decimation filter. During playback, the sound is generated simply by passing it through an analog LPE.
This is a high-speed sampling 1-bit ΣΔ modulation recording method proposed by Professor Yamazaki of Waseda University.
Nagae: Waseda University's high-speed Hashimoto-type 1-bit ΣΔ converter has a 7th-order characteristic as shown in Figure 6, but I suspect Sony Philips' DSD has a 5th-order characteristic as shown in Figure 7.
Professor Yamazaki's high-speed sampling 1-bit method used emphasis and the sampling frequency was a multiple of 48 kHz, but it may now be a multiple of 44.1 kHz.
Koban: So the 1-bit ΣΔ modulation times are different between Sony and Hayakada University.
Nagae: Yes, the basic idea is the same as Professor Yamazaki's high-speed sampling 1-bit ΣΔ modulation recording system.
I'm very interested in why Sony and Philips decided to adopt this high-speed 1-bit method.
(Actually, this is a secret, but I know that Anazawa Takeaki of Nippon Columbia, who was conducting digital recording in Europe using the world's first practical PCM recorder, met with Gall of Philips in the Netherlands to discuss the next-generation Super Audio standard. I also know that N, who was Anazawa's subordinate at the time, reported that a 1-bit DSD signal could be obtained from Waseda University's high-speed sampling 1-bit method and pins 8 and 21 of the AK-5390 IC from Asahi Kasei Microsystems, which was commercially available at the time.)
Ko: Perhaps someone proposed this A/D conversion method.
Nagae: Perhaps this is because people mistakenly believe that DSD is a completely new method different from conventional A/D conversion methods. It uses the first half of the A/D conversion IC found in DAT and MD.
As such, the component itself is mass-produced.
Kotane: If you pass it through a decimation filter in the second half of the A/D conversion IC, it becomes what's known as multi-bit.
Hayami: So that means DSD signals can also be converted to DVD-Audio encoding (National 8).
Oba: Yes. The output bit depth can be any number of bits if you pass it through a filter.
Nagae: The fact that one master sound source can be converted into various formats is what's attractive to record companies.
I think Sony's particular achievement was making the sampling frequency a multiple of CD's 44.1 Hz.
Kago: Mathematical data can also be 32-bit or 64-bit, so it's fine to record it using the DSD method.
Nagae: Why not just go with 1.4MHz/1 million bits? (laughs)
Kakimi: I see! The sampling frequency and quantization bit depth mentioned in DVD-Audio simply refer to how many bits are output from the digital filter!
Obata: There's also a technique called dithering. I don't think there's much point in getting hung up on the number of output bits or sampling frequency of a digital IC.
Nagae: What we should be concerned about is the quality of the high-speed sampling 1-bit ΣΔ modulation circuit.


Calculation Errors Due to Filter Dependence

Obata: This number is similar to that of digital filters in CD players.
As with 8fs/20bit, increasing the number does not necessarily improve the sound quality of the audio source.
Nagae: DATs are equipped with a 16-bit output A/D conversion IC, the AK-5326, from Aika.
A sister model, the AK-5328, has an 18-bit output, but the dynamic range does not become 18-bit.
MJ: But wouldn't a multi-bit output reduce filter calculation errors?
Kago: It's pointless if errors occur in the filter's internal circuitry.
MJ: I heard that the offset becomes looser each time filters are cascaded.
Nagae: The problem with cascading filters isn't offset offsets. The stopband attenuation characteristics become uneven rather than smooth, and errors occur due to error truncation when transferring data between stages.
This is the same as with the digital filters in CD players.
A CD player's digital filter amplifies the input signal in three stages: 2x, 4x, and 8x. Perhaps cascading filters can be effective against this.
MJ: I heard Sony developed a technology called SBM Direct to avoid calculation errors during decimation (Figure 9).
Obata: This time, we'll be talking about the decimation filter in the A/D conversion circuit.
Nagae: This downsamples a 2.8224MHz/1 hit signal to 441kHz in one go using a 32,768-tap filter, and then applies digital ΣΔ modulation for SBM characteristics to the 24-bit data, raising the noise floor in the frequency range with high auditory sensitivity.
Obata: I heard the 2MHz setting is for surround playback.
MJ: Aren't all current A/D conversions cascaded?
Nagae: No. We're also mass-producing a 4,096-tap filter with 18-bit output and a 20-bit output A/D conversion IC with 204 taps.
By the way, surround sound may be used together with video. Since the 5.1ch master for movies is 48kHz, I think it should be a multiple of 48kHz, but what will actually happen?
Koda: If you choose a sampling frequency based on 44.1kHz, you might get complaints from Hollywood.


Dream or reality? 200kHz playback

MJ: DSD now allows recording and playback up to 100kHz.
Hayami: From the perspective of waveform reproduction, I think it makes sense to increase the sampling frequency. The DSD technology introduction also uses a 10kHz square wave as an explanation.
Obata: "A 10kHz square wave becomes a sine wave at the CD sampling frequency of 44.1kHz.
Even DVD-Video's 96kHz approaches a sine wave.
However, with a sampling frequency of 28224MHz,
DSD makes it possible to play back 10kHz square waves."
Nagae: It used to be much rarer, but these days, S/N ratios are being renumbered as if they were written by hand in pencil. (laughs)
MJ: It's said that 192kHz was added to DVD-Audio due to the influence of SACD's 100kHz playback.
Nagae: With DSD, the playback bandwidth is determined by the cutoff frequency of the analog LPE on the playback side, so 200kHz playback is easily supported.
Obata: So the disc records up to 1.4MHz.
Fukami: When SACD advertised 200kHz playback, will DVD-Audio offer a 384kHz/24-bit mode?
MJ: An additional interface standard would be necessary.
Fukami: 96kHz-compatible DACs have finally been released, and we're on the verge of 192kHz, so are the standards changing again?!
Obata: I think the internal player transmission bitrate was limited to 9.6Mbps for two channels. It's also 18.432Mbps, so 384kHz/24-bit is impossible.


Recorded on an Optical Disc

Kobashi: It seems that passing data that falls far short of 24-bit performance through a 24-bit output decimation filter results in a 24-bit A/D conversion circuit, but what is its actual performance?
Nagae: It depends on whether the bandwidth is set to 20kHz, 40kHz, or 300kHz.
Kotane: Even if harmonic components in music are present, their amplitude is not large, so isn't allocating 20 bits of memory to that band a waste?
Nagae: If you record too long on a high-quality disc, the copyright fees alone would be prohibitively expensive. That's why we record about 80 minutes.
Kotane: I'd like to see a reasonable waste of memory that matches the performance.
Nagae: It would be counterproductive to allocate too much recording capacity above 20kHz, resulting in insufficient memory on the optical disc and having to introduce lossless compression.
I think memory should be allocated to bands with high auditory sensitivity.
Senmi: In the demo sound source at Waseda University,
in the frequency range above 20kHz, what percentage was quantization noise, and what percentage was musical harmonic components?
Nagae: It depends on the quality of the A/D converter, but even if more than 99% was quantization noise, it was inaudible.


Discussion of Range Using FFT Measurement Instruments

Nagae: When the 2.8224MHz output signal of a certain high-performance 1-bit A/D conversion IC is passed through a 100kHz analog LPF, the spectrum shown in Figure 10 is obtained.
Obata: There's quite a lot of residual quantization noise.
Nagae: But this actually sounds quite good. Even with the inclusion of large-amplitude noise, it's not at all noisy.
Obata: This is an amazing waveform that makes faithful waveform reproduction seem ridiculous.
Nagae: Noise can be completely removed using a digital filter. (Figure 11)
Obata: The cutoff characteristic of an analog LPF isn't very steep.
Hasumi: So you're saying that the inclusion of large-amplitude high-frequency noise results in a better sound? What exactly does that mean?
Nagae: It's not that the sound is good because it contains high-frequency components; the sound quality deteriorates when high frequencies are cut.
The problem is that multiple cascaded finite-word-length digital filters are used when converting from AD to DA. Word length truncation and calculation errors only have an effect in bands with high audible sensitivity.
Analog LPFs do not have calculation errors from word length truncation, so distortion does not occur when calculation errors occur.
Obata: Some audiophiles might frown when looking at a 10kHz square wave on an oscilloscope.
Hasumi: What does it mean that even noise is inaudible? Was the sound recorded above 20kHz quantization noise, or was it a musical component?
Nagae: Did you hear it? Did you feel it? Did it feel good? These are all subjective questions, so we don't know, but it does prove that D-range is not necessary for high frequency bands.
Obata: At a briefing on a high-speed sampling 1-bit method, an engineer gave an impassioned speech, saying, "FFT measurements showed that our company's A/D conversion circuit has such high performance."
Nagae: "Our company," you say? Isn't that Waseda University's 7th-order characteristic modulation circuit? (Figure 6)
Obata: "How many points does the FFT have?" I asked.
It seems they didn't understand my point, which was, "If you increase the number of points in the FFT, noise outside of 1 kHz will appear lower, so you won't be able to tell how much dynamic range there is."


Trends in Σ-Δ A/D Converters

Nagae: By the way, did you know that 1-bit Σ-Δ A/D converters have become less common recently?
Obata: Huh? Is that true?
Nagae: The reason for this is that 2-level (1-bit) converters cannot use dithering, which can cause the Σ-Δ modulator's loop filter to oscillate, and the need for a high-order Σ-Δ modulator increases out-of-band quantization noise.
MJ: So, does this mean that the mainstream of high-performance A/D converters is moving toward multi-bit Σ-Δ converters?
Nagae: That's correct. Multi-bit Σ-Δ converters themselves were announced and put into practical use 15 years ago. Japan's Asahi Kasei Microsystems, as well as America's Analog Devices, Burr-Brown, and Cirrus Logic are also using multi-bit Σ-Δ modulators.
Higher-order Σ-Δ modulators result in significant noise above 20 MHz, so they're moving toward multi-bit Σ-Δ converters to maintain a consistent S/N ratio.
Koba: However, not just any multi-bit ΣΔ type will do.
It's important that the loop filter design is excellent and stable.
Nagae: Yes. That's right. The key is to avoid oscillation and the "beep" sound that occurs when an offset is added.
Hayami: What about a 1-bit A/D converter for DSD?
Nagae: I don't know. We need to use a conventional IC, turn Waseda University's circuit into an IC, or develop an entirely new architecture. But even if we use a conventional IC, as long as we don't push the frequency characteristics too hard, we can achieve the sound quality that is characteristic of Sojutsu.


Beware of quantization noise

Nagae: If someone shows you an FFT graph and says, "The A/D conversion circuit we developed is this high-performance," why not request, "I want to experience that excellent dynamic range, so please play the opening part of Bolero on the label at full volume?"


https://youtu.be/cmNEvSFWftc?si=6gJrldqXGuSLmfpi


Obatai: So this tests for minute levels of quantization noise, which is a problem with undithered CDs?
Nagae: Yes. It's certainly a quantization noise test, but its main purpose is to ensure the safety of the super tweeter.
Hasumi: Huh?
Nagae: Could tragedy strike if quantization noise is insufficiently blocked?
MJ: I don't really understand what you're saying.
Nagae: Modern super tweeters can reproduce up to 100 kHz, right?
Won't the tweeter's voice coil melt if you turn up the amp volume thinking you can't hear it?
Everyone: I see!
Hasumi: What will you do if it breaks? Some super tweeters cost as much as 800,000 yen each! Nagae: The cutoff characteristics of the playback LPF should be such that the tweeter won't be damaged.
I think it's best not to be greedy with the frequency characteristics.
Obata: You need to measure things like, "To what extent should fundamental noise be suppressed during playback?" and "What kind of analog LPF circuit is needed?"
MJ: You can also consider various filter circuits. Passive LC type, active LPE, etc.
Nagae: Be careful of the coil's nonlinearity. Be careful of OP amp TIM distortion.
MJ: Super Audio is a super analog filter.
Nagae: You'll get better results by gradually attenuating the LPF cutoff frequency from around 30kHz rather than extending it all the way to 100kHz.
Hasumi: But if you think about it, SACD 6 and DVD-Audio 6 won't be released without editing like live recordings, right?
Nagae: Oh! That's right!
Obata: Distortion will be added due to calculation errors in the editing machine. Quantization noise may also be cut out on the editing machine.
Nagae: If you cut out the quantization noise, the recording-to-playback characteristics won't be 100kHz, but isn't that a bit disappointing?


Future Challenges

Kago: When formulating a new standard, we'll also need a vision for "high-data-rate A/D conversion circuits" and "a vision for computational errors in digital editing machines."
Nagae: It's about time to verify the effectiveness of wideband reproduction. We need to consider what frequency range should be reproduced and band-limiting circuits that minimize sound quality degradation.
It's problematic to compete over numbers without considering the actual A/D conversion circuit, or to discuss frequency bands based on mood.
Hasumi: We'd like a unified standard that we can recommend with confidence.
At the very least, we'd like a standard that will ensure that super-audio players won't break even when listening at high volume.
MJ: Thank you very much for your time today.















Daisuke Obata:
Born in Tokyo, age 43.
TEDDIGITAL International Consulting

Tetsuya Nagae:
Born in Tokyo, age 43.
Works for an audio equipment software company, that is DENON Nippon Columbia Ltd.
Currently in the multimedia department.
Formerly worked in audio equipment product planning.
He has connections with semiconductor and audio equipment manufacturers around the world.

Hisashi Hasumi:
Born in Tokyo, age 45.
Works at IKEON, a high-end audio shop in Ikebukuro, Tokyo.
His weekly Friday high-end audio equipment listening sessions are well-known in the industry.
As of August 1998, 380 listening sessions had already been held.