Analog Devices AD1139(richis-lab.de)
richis-lab.de
Analog Devices AD1139
https://www.richis-lab.de/DAC07.htm
42 comments
Really stupid question: I thought I've used a 24bit DAC, on some high end sound equipment. Why is this special/noteworthy?
Not all DACs are equal.
Even average soundcards sport a 24bit DAC. However they're designed for a dynamic output so tend to have quite crap DC precision. What matters to a soundcard is producing a precise waveform, not a precise DC voltage.
This part is designed for DC level setting, ie a programmable reference voltage. Typically it won't change that often (compared to a soundcard), but it needs to be precise.
For example, compare the AD5791[1], a 20-bit DAC with 1LSB DC accuracy costing $40 in 1000 qty, to a 24bit audio DAC like the AD1955[2] costing about $7. The datasheet for the audio DAC doesn't even specify the DC precision, only a 6% gain error from the output buffer.
[1]: https://www.analog.com/en/products/ad5791.html
[2]: https://www.analog.com/en/products/ad1955.html
Even average soundcards sport a 24bit DAC. However they're designed for a dynamic output so tend to have quite crap DC precision. What matters to a soundcard is producing a precise waveform, not a precise DC voltage.
This part is designed for DC level setting, ie a programmable reference voltage. Typically it won't change that often (compared to a soundcard), but it needs to be precise.
For example, compare the AD5791[1], a 20-bit DAC with 1LSB DC accuracy costing $40 in 1000 qty, to a 24bit audio DAC like the AD1955[2] costing about $7. The datasheet for the audio DAC doesn't even specify the DC precision, only a 6% gain error from the output buffer.
[1]: https://www.analog.com/en/products/ad5791.html
[2]: https://www.analog.com/en/products/ad1955.html
On the other hand if all you care about is DC precision you can use a very simple, though not necessarily easy, circuit. You need a stable voltage derived from a stable voltage reference, and a stable clock to PWM that stable voltage. Then you "just" low-pass filter the heck out of the PWM to get an adjustable DC value, and calibrate out any offsets. I've seen this used in adjustable voltage standards (forget which) and the venerable Fluke 5700A Multifunction Calibrator[0].
[0]: https://www.youtube.com/watch?v=qhr6h7wuqo0
[0]: https://www.youtube.com/watch?v=qhr6h7wuqo0
Most "24 bit" audio DACs have a lot less than 24 bits of real usable data due to the noise floor effectively burying the lowest bits in garbage, and I think the noise increases at less perceptually noticable frequencies too. 24 bits is used for audio mainly because it's just convenient to standardise on that. I think they also have a lot more non-ideal behaviour in general than this, and they certainly don't have a 0.1% precision voltage reference because the precise output voltage range doesn't really matter for audio.
I doubt anyone ever shipped this DAC for digital audio applications. You're right that on sheer width there have been 18-bit or better DACs on the market since the 1980s. First one that comes to mind is the Burr-Brown (now Texas Instruments) PCM58. ADI themselves have been marketing 18-bit or better audio DACs for ages, since at least the AD1860. Compare the datasheet of the AD1860 to this one to see the difference between a precision DC instrument and a dynamic audio DAC. This one has .01% gain error while an audio DAC is rated for 2% gain error (because nobody cares).
These industrial DACs are actually brilliant for audio - They retain low-end and transients much better, but it also means need to be calibrated properly so that DC don't destroy speakers. It's easier and cheaper to just add HPF though, but sounds "worse".
You can specify 144 dB of dynamic range but in practice:
1) Loudness war means pro sound engineers will mix everything to be full scale all the time so music has 0 dB of dynamic range ideally
2) Almost no one owns equipment capable of rendering 144 dB of dynamic range even if they have a bit stream specifying it. So if a quiet near silent room is 30 dBA where you hear your own heartbeat and OSHA permissible exposure limit is 90 dBA its only 60 dB of dynamic range between "too quiet to hear" and "so loud you legally need earplugs" so you'll be throwing out about 84 dB of theoretical dynamic range anyway even if you owned gear capable of rendering it. Note that its very hard to buy audio amplifiers capable of more than 90 dB or so SNR even if just hooked up to test equipment, nothing on the market capable of 144 dB, so either the bottom end will be lost in the hiss or the high end will be distorted into unrecognizability.
3) Then you get into topics like power supply rejection ratio. A "bad" amp would pass noise from the power supply thru. A "better" amp rejects more noise from the power supply. A good way to look at it is the hummmmm and crackle and stuff that you hear on a bad old tube amp could be modeled as a very large value attenuator directly connecting the power supply to the output. With 1970s tech and capacitors, 80 dB was considered pretty good PSRR. The very top end newest class D amps can just barely exceed 100 dB PSRR under idealized testing conditions. So at full blast (which would just blow your eardrums out anyway) everything more than 100 dB down from peak will merely be power supply noise passed thru the system.
Its kind of like I can spec a piece of wood to 1.5875mm thick. And then convert it to imperial inch measurements. If the best measurement tool I have to measure wood, is a carpenters tape measure ruled to eighths of inches, the best measurement I can do to render those five digits of precision is "about half a division" on that tape measure. The five digits of precision in the mm figure imply I should be using a calibrated micrometer to measure that thickness, but if you assume technology hasn't invented them yet, then there's little point in spec a digital calculated measurement to five digits of mm precision. Likewise I can make a string of binary 1 and 0 that theoretically can be understood or mathematically proven to imply 144 dB of sound precision, but even in 2021 electronics hardware doesn't exist to reliably repeatedly and provably render those 1 and 0 into sound, at least not at audio rates.
Note that 2 to the power of 24 is a eight digit number, implying you're operating at tens of ppb. NIST has really nice new Josephson junction standards capable of reliably operating at tenths of ppb. So you're very optimistically trying to field gear operating at only a hundred times less accurate than the best lab in the entire world can currently measure. Good luck with that. That's why nobody sells voltmeters with more than 8 or so digits of precision, if the world standard calibration system only has ten digits on its best days, no point building voltmeter hardware displaying 12 or 15 digits LOL. We don't really have the technology to do 24 bit accurate voltages out in the field right now. So if we can't build or calibrate testing gear significantly better than 24 bits, like to 32 or so bits perhaps, there's no way to actually measure and tell if 24 bit gear is accurately precisely repeatedly working to 24 actual bits of precision. Maybe your audio gear is distorting in a subtle fashion and only operating to 22 noise free bit and the last two bits are essentially a RNG or stuck on/off or otherwise no relationship with reality; very few electronics labs in the entire world have the gear to prove that claim true or false.
1) Loudness war means pro sound engineers will mix everything to be full scale all the time so music has 0 dB of dynamic range ideally
2) Almost no one owns equipment capable of rendering 144 dB of dynamic range even if they have a bit stream specifying it. So if a quiet near silent room is 30 dBA where you hear your own heartbeat and OSHA permissible exposure limit is 90 dBA its only 60 dB of dynamic range between "too quiet to hear" and "so loud you legally need earplugs" so you'll be throwing out about 84 dB of theoretical dynamic range anyway even if you owned gear capable of rendering it. Note that its very hard to buy audio amplifiers capable of more than 90 dB or so SNR even if just hooked up to test equipment, nothing on the market capable of 144 dB, so either the bottom end will be lost in the hiss or the high end will be distorted into unrecognizability.
3) Then you get into topics like power supply rejection ratio. A "bad" amp would pass noise from the power supply thru. A "better" amp rejects more noise from the power supply. A good way to look at it is the hummmmm and crackle and stuff that you hear on a bad old tube amp could be modeled as a very large value attenuator directly connecting the power supply to the output. With 1970s tech and capacitors, 80 dB was considered pretty good PSRR. The very top end newest class D amps can just barely exceed 100 dB PSRR under idealized testing conditions. So at full blast (which would just blow your eardrums out anyway) everything more than 100 dB down from peak will merely be power supply noise passed thru the system.
Its kind of like I can spec a piece of wood to 1.5875mm thick. And then convert it to imperial inch measurements. If the best measurement tool I have to measure wood, is a carpenters tape measure ruled to eighths of inches, the best measurement I can do to render those five digits of precision is "about half a division" on that tape measure. The five digits of precision in the mm figure imply I should be using a calibrated micrometer to measure that thickness, but if you assume technology hasn't invented them yet, then there's little point in spec a digital calculated measurement to five digits of mm precision. Likewise I can make a string of binary 1 and 0 that theoretically can be understood or mathematically proven to imply 144 dB of sound precision, but even in 2021 electronics hardware doesn't exist to reliably repeatedly and provably render those 1 and 0 into sound, at least not at audio rates.
Note that 2 to the power of 24 is a eight digit number, implying you're operating at tens of ppb. NIST has really nice new Josephson junction standards capable of reliably operating at tenths of ppb. So you're very optimistically trying to field gear operating at only a hundred times less accurate than the best lab in the entire world can currently measure. Good luck with that. That's why nobody sells voltmeters with more than 8 or so digits of precision, if the world standard calibration system only has ten digits on its best days, no point building voltmeter hardware displaying 12 or 15 digits LOL. We don't really have the technology to do 24 bit accurate voltages out in the field right now. So if we can't build or calibrate testing gear significantly better than 24 bits, like to 32 or so bits perhaps, there's no way to actually measure and tell if 24 bit gear is accurately precisely repeatedly working to 24 actual bits of precision. Maybe your audio gear is distorting in a subtle fashion and only operating to 22 noise free bit and the last two bits are essentially a RNG or stuck on/off or otherwise no relationship with reality; very few electronics labs in the entire world have the gear to prove that claim true or false.
I believe this chip was first available in 1989. It's old technology.
Are the traces gold plated?
Yep. This hybrid was probably manufactured in open air, so they would want gold plating to prevent it from oxidizing while heated during wirebonding / die attach.
Really nice tear down. These digital synthesizers have been really great for breathing new life into amateur radio and digital single generation.
Beautiful pics. I wonder why, in a few places, there are pairs of bonding wires connecting the same pads.
> Beautiful pics.
He uses[1] Helicon Focus[2] for focus stacking.
A lot of these die shots are pieces of art, and also quite interesting. I especially like the LTZ1000 voltage reference[3] or the Thyracont Inclinos[4] which is an accelerometer without moving parts. A lot of the MEMS dies are also very interesting, because of their moving parts, like this[5] ADXL213.
[1]: https://www.eevblog.com/forum/projects/transistors-die-pictu...
[2]: https://www.heliconsoft.com/heliconsoft-products/helicon-foc...
[3]: https://www.richis-lab.de/REF03.htm
[4]: https://www.richis-lab.de/MEMS_04.htm
[5]: https://www.richis-lab.de/ADXL213.htm
He uses[1] Helicon Focus[2] for focus stacking.
A lot of these die shots are pieces of art, and also quite interesting. I especially like the LTZ1000 voltage reference[3] or the Thyracont Inclinos[4] which is an accelerometer without moving parts. A lot of the MEMS dies are also very interesting, because of their moving parts, like this[5] ADXL213.
[1]: https://www.eevblog.com/forum/projects/transistors-die-pictu...
[2]: https://www.heliconsoft.com/heliconsoft-products/helicon-foc...
[3]: https://www.richis-lab.de/REF03.htm
[4]: https://www.richis-lab.de/MEMS_04.htm
[5]: https://www.richis-lab.de/ADXL213.htm
I’m speculating but when I’ve seen it in the past it was for handling higher current.
In this case I’m wondering if it’s used to maintain better control of the voltage gradient on a pad.
In this case I’m wondering if it’s used to maintain better control of the voltage gradient on a pad.
Sometimes it's to handle higher currents, other times it's to reduce the bond wire inductance and improve the high-frequency response. For power pins it can be both. Especially for a multi-chip module with some digital circuitry, there's potential for high-frequency digital transients that need nearby decoupling capacitors through low-inductance loops, or the transistor switching noise can pollute other copackaged circuit supplies; amplifiers also benefit from lower inductance on the supply lines since it stabilizes the frequency response at higher offsets beyond the loop bandwidth of any integrated supply regulation, current sources, bandgap references etc. So you're essentially correct, it's about voltage gradients, but probably at higher frequency.
AD make some good chips. Along with Realtek they kickstarted the SDR revolution. No hype or trite SaaS products here, just good old school engineering.
If only they paid like they sold ads.
If only they paid like they sold ads.
it annoys me to an unbelievable levels that all that CRUD SaaS crap is called "tech" nowadays.
We're veering pretty far off-topic, but, as a mechanical engineer working in research, yes.
Develop a new ceramic material for advanced turbofan engines? Not tech.
Web front end? Oh yeah, tech.
Develop a new ceramic material for advanced turbofan engines? Not tech.
Web front end? Oh yeah, tech.
Isn’t that just called engineering?
No, technology is the application of science and other knowledge to achieve practical goals. It started with agriculture, metallurgy, etc. thousands of years ago.
Your answer is very condescending.
I am referring to startup industry terms. Tech for most web related activities, engineering or Deep tech for the others mentioned in this thread.
Excerpt from the wiki[0]:
> Deep (or hard) tech is a classification of an organization, or more typically a startup company, with the expressed objective of providing technology solutions based on substantial scientific or engineering challenges.
[0]: https://en.wikipedia.org/wiki/Deep_tech
I am referring to startup industry terms. Tech for most web related activities, engineering or Deep tech for the others mentioned in this thread.
Excerpt from the wiki[0]:
> Deep (or hard) tech is a classification of an organization, or more typically a startup company, with the expressed objective of providing technology solutions based on substantial scientific or engineering challenges.
[0]: https://en.wikipedia.org/wiki/Deep_tech
It’s annoying for sure. But, in fairness to those SAAS apps, ”tech”, as recognized by the mass audience, is short for information technology. Meaning conveying knowledge to another person as efficiently as possible. So that lame-sounding SAAS app might really be a small leap forward in transfer of knowledge. Even if the underlying implementation is something a high school student could figure out.
But think of the epic problems they needed to solve to get the protocol implemented using websockets in javascript so it can run in the iframe on the mobile browsers! Such engineering prowess! </sarcastic>
Great tear down but I’m more of a TDA1541 man myself.
Some context:
>The AD1139 is the first DAC offering 18-bit resolution (1 part in 262,1441) and true 18-bit accuracy in a component size hybrid package. A proprietary bit switching technique provides high accuracy, speed and stability without compromising small size or low cost.[0]
Just curious, I looked up the price:
Digikey, 1 unit: $2,449.80 USD [1]
Analog Devices, 100-499 units: $1706.58 USD
Analog Devices, 1000+ units: $1706.58 USD
[0]: https://www.analog.com/en/products/ad1139.html
[1]: https://www.digikey.com/en/products/detail/analog-devices-in...
>The AD1139 is the first DAC offering 18-bit resolution (1 part in 262,1441) and true 18-bit accuracy in a component size hybrid package. A proprietary bit switching technique provides high accuracy, speed and stability without compromising small size or low cost.[0]
Just curious, I looked up the price:
Digikey, 1 unit: $2,449.80 USD [1]
Analog Devices, 100-499 units: $1706.58 USD
Analog Devices, 1000+ units: $1706.58 USD
[0]: https://www.analog.com/en/products/ad1139.html
[1]: https://www.digikey.com/en/products/detail/analog-devices-in...
There seems to be no price cap on AD components! Solid stuff at a range of prices. It's sometimes tough to tell what makes the super-expensive ones different. Might be a little extra precision (etc) that's justified in niche cases.
Isn't is just market segmentation? The only people who need this thing are building test equipment that costs big money regardless.
I'd love to have one in my radio
How could one use an 18-bit D/A in a radio? (an 18-bit A/D is a different story).
I'd say volume. These are highly specialized devices which require a lot of careful development. The volume might be a few thousand a year if even. So seeing super high speed/precision ADC's costing 1000+USD isn't so surprising.
that's interesting, ADI doesn't even list it anymore on their website but it's marked as still active. i guess it's still used in some test&measurement equipment.
this feels like something to throw into a satellite for processing/digitizing the output from a custom image sensor?
or maybe some incredibly edge software-defined radio application...
or maybe some incredibly edge software-defined radio application...
Great write up - thanks OP, I'd not seen a hybrid torn down like that before.
When was this released?
Looks like it's as least as old as the 90s. Found a picture of one sporting a 9943 date code. The one in the article is from 2002.
Analog still lists them in production!
Analog still lists them in production!
The datasheet PDF doesn't contain an obvious date, but it's is clearly scanned and it also contains a sample circuit for connecting it to an IBM PC via the ISA bus. So somewhere between 1981 and 1995, perhaps.
There's a document id on the last page that may indicate it's from 89.
https://www.analog.com/media/en/technical-documentation/data...
There's a document id on the last page that may indicate it's from 89.
https://www.analog.com/media/en/technical-documentation/data...