Saturday 25 March 2023

RETRO-AUTOPSY: Museatex Melior BS-1 Bitstream S/PDIF DAC (circa 1995?). A few words about external clocks for audiophiles. And a speaker cable blind test?

The company Museatex was formed around 1990 by Ed Meitner and John Wright based in Calgary, Alberta, Canada. Over the years, the brand was assumed under A/D/S which eventually closed in 2001. There's still a web page for information on Museatex products here. Ed Meitner has moved on these days to EMM Labs since 1998 and given his work with SACD, his name has been associated with "bitstream" type technologies since (some more info/video).

By the end of the 1980's, 1-bit Philips "Bitstream" PDM DAC chips started to show up in the market with the Philips SAA 7321 chip as one of the 1st generation devices (see Stereophile reporting back in 1989). Some of these chips were incorporated in CD players like the Meridian 208 that came out in early 1990, well-reviewed by Paul Miller in High Fidelity.

By 1991, the "second generation" Philips SAA 7350 "bitstream" DAC became available, capable of taking up to 20-bit PCM (from 16-53kHz sample rate) and upsampling to 1-bit SDM applying 3rd order noise shaping (some more history here). It is this DAC chip that's at the heart of the Museatex Melior Bitstream DAC, I believe first released in 1992. Here are the specs from the company.

The exact history of this particular unit I had access to has been shrouded in the sands of time, but was probably purchased around 1995. The original suggested price on one of these was US$899 which would be about equivalent to over US$1500 in 2023. Even to this day, I've seen comments from audiophiles who have a special affinity to the sound of this vintage DAC. Claims to fame included anti-jitter input circuitry (supposedly reducing "glare and harshness"), a custom digital filter allowing for "more accurate signal conversion", and high quality power supply with line noise filtering.

Thanks again to my friend linnrd for digging this unit out from his audio device archives. While externally it looks to be in very good shape, unfortunately, it didn't quite power up normally for me which sadly resulted in this article being more of an "autopsy" rather than actual measurements. I guess having a look inside is still better than nothing for historical documentation if nothing else.

Saturday 18 March 2023

REVIEW/MEASUREMENTS: AOSHIDA A7 Class D TI TPA3255 amplifier. Good sound. Good power. Economical.

Above, you see the small AOSHIDA A7 Class D amplifier which we'll go in depth with in this post. It's inexpensive and can be purchased for less than US$150 these days. This design is also sold as the Douk Audio H7 with a different color scheme, presumably the innards must be very similar if not identical. I preferred the look of the black AOSHIDA version with orange-ringed VU meter. I purchased this amplifier through the usual retail channels.

Over the years, it has been good to see the evolution of Class D designs and the inexpensive "chip" amplifiers. Internally, this is based on the TI TPA3255 "PurePath" chip (released in 2016) rated to provide up to 315Wpc into 4Ω stereo but at 10% THD when provided with adequate power (data sheet max is 53.5V). Notice the italics to emphasize that these numbers must be put into context. For audiophiles who want lower distortion in their amps (like <1% THD), we would be looking at much more conservative power figures than that 300+W being promoted for these amps.

(The chip also has a mono configuration that's spec'ed at up to 600W into 2Ω which would not apply for this review.)

Saturday 11 March 2023

The Audiophile Wisdom of ChatGPT (Feb 13, 2023 Version): Learnings from an artificial intellect on an esoteric hobby. (And on creativity?)

I'm sure none of us these days have escaped news about ChatGPT from OpenAI. Articles have been ubiquitous covering the implications of what this "chatbot" is able to do. No doubt, there will be deep implications stemming from this technology already being debated through the halls of academia on how it will affect everything from education, to work, to legal ramifications. As with any likely transformative technology, there will be promise and fear.

Already, I've seen many articles out there including in our little corner of the universe with Stereophile's "As We See It" in February referencing ChatGPT.

As you're likely aware, ChatGPT achieves its remarkable abilities based on Generative Pre-trained Transformer (GPT-3.5) "large language model", a sophisticated "self-supervised" general AI model based on >175B parameters combed through 570GB of data as text off the Internet, books, and Wikipedia. Yes, human feedback is part of the training process and data continues to be gathered from users to improve the model. At the time of this article, the openly available online ChatGPT has been trained with data up until September 2021 as can be seen here:

Arch: Who won the Superbowl in 2023?

ChatGPT: I'm sorry, but I am not capable of predicting future events as I am a machine learning model trained on historical data. As of my knowledge cutoff date, which is in September 2021, the most recent Super Bowl was Super Bowl LVII and it has not happened yet. It is scheduled to take place on February 5, 2023, so we won't know the winner until the game is played.

Given the level of knowledge ChatGPT has access to, I wondered how would it perform currently in March 2023 if asked about broad audiophile topics. I think this is interesting because the answers it gives must be an integration of all those hundreds of gigabytes of texts fed into the system. What does all that information teach the machine about the current understanding of audiophile topics? Are the answers reasonable, and relatively free of errors? Do the answers show "common sense" or even a level of "wisdom" one might wish to receive when talking to a knowledgeable audiophile friend requesting advice?

Let's dive into a Q&A discussion and see what ChatGPT can tell us about this niche audiophile hobby... (Grab a warm beverage by the fireside friends, we go deep and broad.)

Saturday 4 March 2023

Internet Blind Test: 24-Bit vs. Dithered 16-Bit Part Deux - Daft Punk Edition! And the honest desire to seek truth as audiophiles.

BTW: 1.5kHz sine waveform, -65dBFS. 16-bits on the left, 24-bits on the right.

It's amazing how time flies. Nine years ago - 2014 - at a time when hi-res audio downloads were being promoted as the "next best thing" for audiophiles, we ran a 24-bit vs. 16-bit blind listening test on this blog. I'll leave you to examine the results.

Since then, electronics like DACs have advanced quite significantly. Compared to 2014, these days, high quality asynchronous USB DACs have become ubiquitous. Jitter is basically a non-issue now. Inexpensive audio streamers like Raspberry Pi devices can be easily put together. We can all play music with wide selections of digital filters without great difficulty. Very high quality headphone amps and phenomenal headphones have become commonplace with a growing "head-fi" community.

Recently, I had some fun on this Steve Hoffman Forums thread engaging in the discussion about 16-bit vs. 24-bit; the topic of audibility again came up. As usual, things can get pretty contentious among audiophiles and since it has been awhile and perhaps some of you did not have the opportunity to participate in 2014, let's do a blind test again for those wanting to listen to 24-bit vs. 16-bit audio.

Let's use popular music rather than the more esoteric 24-bit 2L classical recordings last time - let's try some "classic" Daft Punk. ;-)