|
Post by jeromemason on Jul 8, 2018 0:03:25 GMT -6
svartI know you already went down this rabbit hole when you made the Svart Box, and I commend you for taking that on. I know a lot of folks around here have enjoyed the benefits of it! But, I really would love to get into the converter side of designing and building, but, I just don't know enough about it. So many variables, so many small details that can totally make or break these designs. I've done well designing front and back end analog circuits, I've made myself boxes that no one else on the planet has and I designed them just for me, just for my workflow, just for my taste. I was over looking at what Lavry is putting out and noticed that put an emphasis on their new (modified version of the AD122-96MkIII) AD122-96 MX being particularly good at overdriving and clipping. Here is their description of it: Lavry AD122-96MX
I've been looking at AKM's newest A/D chips and I've read on other forums of folks using these chips in a parallel design to get crazy specs on dynamic range, I think one I read had one spec'd at something like 150db some higher. With the 32bit chips and some using ultra clean/transparent transformers they're getting some pretty damn amazing performance. Some of these new high quality converters are taking a route of having no electronics whatsoever in the signal path right into the A/D chip. I'm sure you have a full plate svart, if not it'd be cool to start a discussion on something like this, but if you know of any decent resources that I can study to help me a long in starting down this road I'd appreciate it. I really would like to understand more about the DSP side of things, being able to program onboard micro controllers etc. I've read some are using DSP for their clocking to improve effects of temp and eliminating the need for oven controlled crystals etc. But, I would definitely love to embark on throwing ideas around for an A/D that had some cool features to it like tape/transformer efx, being able to clip the converter in a good way. The Lavry box I linked to above, it sells for over $8k..... it'd be awesome to DIY the same quality can performance. Hopefully some good discussion will come of this, maybe even a box! Jerome
|
|
|
Post by svart on Jul 8, 2018 7:49:08 GMT -6
svartI know you already went down this rabbit hole when you made the Svart Box, and I commend you for taking that on. I know a lot of folks around here have enjoyed the benefits of it! But, I really would love to get into the converter side of designing and building, but, I just don't know enough about it. So many variables, so many small details that can totally make or break these designs. I've done well designing front and back end analog circuits, I've made myself boxes that no one else on the planet has and I designed them just for me, just for my workflow, just for my taste. I was over looking at what Lavry is putting out and noticed that put an emphasis on their new (modified version of the AD122-96MkIII) AD122-96 MX being particularly good at overdriving and clipping. Here is their description of it: Lavry AD122-96MX
I've been looking at AKM's newest A/D chips and I've read on other forums of folks using these chips in a parallel design to get crazy specs on dynamic range, I think one I read had one spec'd at something like 150db some higher. With the 32bit chips and some using ultra clean/transparent transformers they're getting some pretty damn amazing performance. Some of these new high quality converters are taking a route of having no electronics whatsoever in the signal path right into the A/D chip. I'm sure you have a full plate svart, if not it'd be cool to start a discussion on something like this, but if you know of any decent resources that I can study to help me a long in starting down this road I'd appreciate it. I really would like to understand more about the DSP side of things, being able to program onboard micro controllers etc. I've read some are using DSP for their clocking to improve effects of temp and eliminating the need for oven controlled crystals etc. But, I would definitely love to embark on throwing ideas around for an A/D that had some cool features to it like tape/transformer efx, being able to clip the converter in a good way. The Lavry box I linked to above, it sells for over $8k..... it'd be awesome to DIY the same quality can performance. Hopefully some good discussion will come of this, maybe even a box! Jerome This will take a lot of discussion, so I'll have to reply tomorrow when I get back into town. Short version is any adc is "clippable", it's just going to come down to how it sounds, how much you can push it, and how the sound breaks when it does. My converter was/is euphoniously clippable to a degree, and I can get around 6db of transient shaving before the digital harshness starts to be evident. That's more than enough, and more than most converters. There's actually a thread around here on transient shaving with audio files I did comparing about 4db of a/d clipping on a track with the raw track and then level matched. People had mixed results picking the clipped track out, and some preferred it.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jul 8, 2018 8:27:02 GMT -6
Be interesting to do that same test with ITB digital clippers and limiters too. Good luck with your project! I rarely clip my Crookwood ADC, only for HipHop clients who want the crazy loud, but I can usually only get around 2-3dB max before it starts sounding like total arse.
|
|
|
Post by jcoutu1 on Jul 8, 2018 8:35:15 GMT -6
|
|
|
Post by jeromemason on Jul 8, 2018 16:15:30 GMT -6
My Lavry AD10 will let me get around 6db before things start to break up. I've thought about pulling it apart and upgrading the opamps and removing some signal caps depending on what the dc readings would be after the new gen chips.
The new AKM chips being 32 bit would give more room to shave transients with the analog front end or if it was a transformer input those would do the same thing, just something like Lundahl, Jensen or Cinemags. Actively I like the way the Opa1642's sound juiced.
The whole reason I'd like to have an A/D that I can clip is mostly so I don't have to use as much buss compression. Clipping the converters is a true 0 attack 0 release so as long as you can't hear any break up it's an extremely transparent limiter.
|
|
kcatthedog
Temp
Super Helpful Dude
Posts: 15,084
Member is Online
|
Post by kcatthedog on Jul 9, 2018 5:28:38 GMT -6
Very interesting and lots of work: given all that work and eventual cost: some reason you are not thinking about the dangerous convert 2+: doesn't it do most of what you want ?
|
|
|
Post by svart on Jul 9, 2018 8:11:12 GMT -6
Ok, so lets chat about this.
So, I see you mentioning 150dB DR, and then mentioning 32bit and transients in the same sentence.. Maybe I need to say that higher bit precision isn't going to get you any different clipping characteristics..
"clipping" is exactly what it sounds like, it's where the signal runs out of headroom, in the analog domain.
This can happen in many places, and in complex ways. The opamps themselves can limit headroom through either input/output structure saturation, or in the case of most modern opamps, their ESD protection structures will start to conduct in an effort to protect the internal parts from being destructively overloaded. Of course, these limits are generally higher than the saturation point of the transistor structures, so the opamp will enter into saturation before the ESD structures go into full conduction.
This can also happen to the input structures of the A/D chip itself. Modern A/D devices will have ESD protection much like the opamps will, and will generally saturate and "bend" the signal to some degree before you get hard digital clipping.
Most of these ESD protection devices are going to be TVS diodes, which are similar to bidirectional zener diodes in that they will conduct and short to ground any signal higher than +/- a specified voltage, but are designed to act extremely quickly.. However, they're still diodes at heart and will conduct a slight amount before they turn on completely.
Another thing to consider is waste heat. All power that is clipped will be turned into heat by the diodes and transistors. As a smart person once said, power is not lost, it's only transformed, something something laws of thermodynamics.. LOL.
So my question to you is:
What exactly are you looking for?
Lavry mentions "settings" for their clipping stage, either 3 or 6 dB. I'm assuming they simply reduce gain or source current in a specific stage or add buffered diodes to achieve this.
You already mention getting around 6dB of saturation before harshness on your converter..
I get about that on my converter..
You're realistically not going to get higher than that until you design a true limiter. At that point you might as well just use the converter full scale without clipping and place a software limiter in your DAW.
Now, for the 150dB.. That's approaching theoretical noise limits, and I can tell you that not even the best instrumentation A/D's are close to that type of usable SNR. I use A/D chips that are 500$+ for a SINGLE input that don't even spec out like that. It sounds more like marketing specs, not true performance, and what they do is measure their devices, and subtract all possible noise sources that aren't their chip to generate the spec. It's a false idol for sure, and doesn't usually even include the quantization noise! Sometimes (read: most of the time) they'll even do things like average the noise, but measure the peak signal which simulates a higher SNR.
What's important is a number called ENOB.. Effective number of bits. This is the real measure of how good an A/D and it's supporting circuits are. This number includes all noise sources in the circuit up to the point of acquisition and measures the usable bits in a perfectly designed A/D system.
Lets say that you have an A/D that's spec'd at 150dB.. But your front-end buffering stage adds 10dB of NF (noise figure) from the active devices, then add the Johnson noise, thermal noise, noise gain, and zener noise from your various active and passive devices, and then account for the theoretical noise power bandwidth of 100KHz, which is -124dBm.. Because the converter is likely spec'd to 150dB at a single, best frequency, and once you start integrating noise power over bandwidth, the theoretical best noise floor is greatly reduced and the BEST you can get is now -124dBm for 100kHz bandwidth. A lot of your active noise will be below the thermal noise floor, and some of the zener noise will be above it.
You've already eaten up any gains you *might* have had from having 150dB vs. 120dB.. If the noise floor is a bit higher, most of those noise sources are washed out and you still get your potential 120dB of SNR, but you still haven't accounted for the noise added by impedance mismatches, ingress from terrestrial sources, residual noise on power supplies, and all other things that add up to ruin your noise floor.
So now you're realistically in the 90-120dB of usable SNR in a perfectly designed and terminated converter. You might have 32 bits, but roughly 25% of those will be lost to noise!
If you're like myself and most every other mix engineer out there you'll also compress and shape the signal until you're working in roughly 5-20dB of average SNR on mixdown anyway, so you're effectively throwing away 80% of your possibly SNR anyway.
So I go back to my question of "what do you want, exactly?"
I know the answer is "150db and clipping", but that doesn't describe your motivation. Are you imagining a large increase in fidelity? You want to effectively limit your signal with almost zero lag?
|
|
kcatthedog
Temp
Super Helpful Dude
Posts: 15,084
Member is Online
|
Post by kcatthedog on Jul 9, 2018 8:18:01 GMT -6
Ok, I admit to loving reading svart’s technical posts! Who else says shit like this, love the “ of course”!! “Of course, these limits are generally higher than the saturation point of the transistor structures, so the opamp will enter into saturation before the ESD structures go into full conduction.”
|
|
|
Post by svart on Jul 9, 2018 8:18:35 GMT -6
It also reminds me of a guitar amp trick called "sagging" where either the pedals or the amp are modified so that one or more of their stages are intentionally starved for current. This allows the signal to pull the power rails a bit and adds saturation.
I wonder if taking a buffer stage and intentionally adding resistors/rheostats in series with the power pins would allow the opamp to start saturation much earlier than it normally would without them.
Other possibilities are stages of biased diodes that allow hand-offs between saturation, but this would lead to a drastic increase in noise.
I wonder if they have JFET stages that limit based on average power. it would be simple to include some kind of buffered averaging circuit that feeds JFETs as active resistors to create limiters, much like the 1176.
|
|
|
Post by svart on Jul 9, 2018 11:26:48 GMT -6
Ok, I admit to loving reading svart’s technical posts! Who else says shit like this, love the “ of course”!! “Of course, these limits are generally higher than the saturation point of the transistor structures, so the opamp will enter into saturation before the ESD structures go into full conduction.” Pretty much everyone in the design world says stuff like this, lol..
|
|
|
Post by wiz on Jul 10, 2018 1:50:23 GMT -6
Damn svart, you saved me all that typing... .... NOT!! 8) great stuff bud Wiz
|
|
|
Post by jeromemason on Jul 10, 2018 12:07:40 GMT -6
Well I think what I truly wanted was to start a discussion, a discussion on how to design an A/D that has some of the features out there that Dangerous and Lavry have built into their new designs, the ability of raising the level without having to use compression or limiters......
Honestly what made me want to travel down this rabbit hole is an album I mixed a little while back that was mastered, by a reputable house, they killed the dynamics of my mix. Not transients, but the dynamics.... They also brought up things I had tucked away, I tucked them away to add a not a translucent texture but a frosted glass texture, bringing them up (forgive my analogies) turned them into stained glass. I've always tried to give the mastering fellows what they want, a file not limited, low in level, high in transient detail but alas, I'm always regretting it when the master comes back.
So, I spoke with a few colleagues here in town, I asked how they approached the ME's to not do this because their masters always sound like the mixes I hear when I ask to listen to the pre mastered version of what I heard on the radio. The answer I always get back is "don't give them any room, get the levels where they would and send that" and they all tell me to invest in one of these A/D's. I'd much rather build one, and build one that has everything I'd want with highest grade components and not $8k! Part of me thinks maybe it's not anything to do with the conversion side, but building a front end and back end to send into the conversion chip and then simply normalizing the printed mix...... It's a good discussion to have as well. I don't think a lot of people understand the whole "they clip their A/D" thing. I knew a lot of it was in the front end and back end, but your post Chris, it did provide a lot of information that I didn't know and I do appreciate you taking time to explain everything.
Hopefully more good things can be brought up and thought up!
|
|
|
Post by svart on Jul 10, 2018 12:18:24 GMT -6
Well I think what I truly wanted was to start a discussion, a discussion on how to design an A/D that has some of the features out there that Dangerous and Lavry have built into their new designs, the ability of raising the level without having to use compression or limiters...... Honestly what made me want to travel down this rabbit hole is an album I mixed a little while back that was mastered, by a reputable house, they killed the dynamics of my mix. Not transients, but the dynamics.... They also brought up things I had tucked away, I tucked them away to add a not a translucent texture but a frosted glass texture, bringing them up (forgive my analogies) turned them into stained glass. I've always tried to give the mastering fellows what they want, a file not limited, low in level, high in transient detail but alas, I'm always regretting it when the master comes back. So, I spoke with a few colleagues here in town, I asked how they approached the ME's to not do this because their masters always sound like the mixes I hear when I ask to listen to the pre mastered version of what I heard on the radio. The answer I always get back is "don't give them any room, get the levels where they would and send that" and they all tell me to invest in one of these A/D's. I'd much rather build one, and build one that has everything I'd want with highest grade components and not $8k! Part of me thinks maybe it's not anything to do with the conversion side, but building a front end and back end to send into the conversion chip and then simply normalizing the printed mix...... It's a good discussion to have as well. I don't think a lot of people understand the whole "they clip their A/D" thing. I knew a lot of it was in the front end and back end, but your post Chris, it did provide a lot of information that I didn't know and I do appreciate you taking time to explain everything. Hopefully more good things can be brought up and thought up! I think the transient clipping works so well because the transients are so fast. A typical drum transient is incredibly fast on the attack to the point where no compressor can react fast enough unless it's a software look-ahead limiter. We can shave those transients off a few dB and nobody even really notices for the most part. It's when we start clipping the "meat" of the signal that we start to hear the distortion more clearly. This is where the signal "breaks" for most converters as the fast transients are barely causing the opamps and protection diodes to react, but the larger and slower RMS signals are causing havoc among the parts that've run out of headroom. I think the most fun idea would be to build a triple buffer, with the middle stage using a dual linear pot as series resistance for the opamp's power rails. You can dial-a-brownout with this and see what happens!
|
|
|
Post by matt@IAA on Jul 10, 2018 12:51:34 GMT -6
You can call it the Bob Sag-it!
|
|
|
Post by jeromemason on Jul 10, 2018 13:49:28 GMT -6
I do like that idea!
Having a couple of different types of parallel feeds for things such as a transistor type harmonics exciter and a Xformer circuit before the “sag section” would be pretty fun to have.
Also, and this is where I’d love to learn more about DSP, but it’d be interesting if there was a way you could have something that brought the peaks to 0db and then had a way to attenuate this.
|
|
|
Post by guitfiddler on Aug 29, 2018 20:57:29 GMT -6
I do like that idea! Having a couple of different types of parallel feeds for things such as a transistor type harmonics exciter and a Xformer circuit before the “sag section” would be pretty fun to have. Also, and this is where I’d love to learn more about DSP, but it’d be interesting if there was a way you could have something that brought the peaks to 0db and then had a way to attenuate this. Sounds like on the Dangerous 2-Bus+ and Convert AD+ blended in one unit with variable Harmonics and X-former options in a dedicated 2 channel converter box? Tweaked to perfection, that is cool! May I add, a couple to three different Transformers options to select from? You will also need an external insert path to add one outboard,(Just saying-why not just go all out) Just a thought.
|
|