For our February Special Report: Semiconductors in Audio, Pro Sound News queried a number of leading design engineers about modern solid state components and their application. Their detailed responses resulted in far more information than we could share in print, so we share the complete range of responses here.
Our dialog was with:
Bowery Engineering Associate
Networked Audio Solutions
Alan T. Meyer
Director of Engineering
Alesis Studio Electronics
Chairman Manifold Labs
President Princeton Digital
V.P. Research & Development
Rupert Neve Designs
Chief Technology Officer
VP of Engineering and Product Development, MI group.
Harman Music Group
Chief Technology Officer
Echo Digital Audio
VP of engineering
A full transcript of their replies follows:
1. For analog design, have you discovered any new chips recently? Are you using different components than you have traditionally?
Marc Lindahl: Yes--THAT corp's THAT4320. Really neat--a complete dynamic processing system with really great performance. They also have an improved version of the line receiver we're using in Plugzilla.
I have stopped using tantalum bypass capacitors - the bane of every designer's existence. Multilayer ceramic cap's have gotten so they have the capacity and are cost effective - and are superior performers.
Michal Jurewicz: We are always on this quest and testing new chips in both the analog and digital domain. This coupled with new circuit solutions allows us to stay within our credo; "our new product must sound at least a notch better than the last generation"
Alan Meyer: Not really, but this is mostly due to our heavy design reuse for analog circuits in our products. Alesis tends to reuse as much as possible in order to improve purchasing power across our entire line of products. However, we have noticed that Texas Instruments has been pushing their line of (Burr-Brown) mic-pre input circuits (INA163, INA217) harder than before. We do like to try and find high quality for a good price, so we try and stay in touch with the major IC manufacturers in case new, low-cost solutions arise.
Tony Agnello: TI PGA. Programmable gain amplifier. They've combined a preamp and programmable gain stage in one IC.
We're using fewer and fewer analog components. The higher levels of integration suck so much functionality into a single chip that design has become a matter of connecting blocks. A key milestone happened nearly twenty-years ago when the first sampling A-to-D converters were introduced, eliminating the need for complex analog antialiasing filters and other analog signal conditioning. The trend has continued but over the last two or three years the level of integration has taken another huge step so that today there dozens of products to chose from ranging from single channel low power converters to multichannel codecs.
Don Elwell: We tend to stick to chips that we've used in the past. We're always hunting for new things that we may want to use in the future. TI has a very good new lineup of analog A/Ds, D/As, and pre-amps that we like, but no new designs incorporating them yet.
B.J. Buchalter: Analog Devices ADG452, ultra-low Rds-On analog switch. This is not really a different type of component, but it provides much better performance than what we had available to us previously.
Kevin Burgin: We are always searching for new components however frequently run into limitations set by the new components. Switching frequencies below 200 kHz interfere with our audio bandwidth. Rail voltages of +5 or +/- 3 volts (and getting smaller every day!) are of little use to us as we are looking for maximum headroom and most analog IC's are 36 volts average maximum. Noise figures and distortion are of importance also. I am always evaluating new components but seldom do they become part of a design unfortunately.
Nathan O'Neill: Our analog audio stages are still predominantly the same: using op-amps for front-end line level inputs, instrumentation amps for mic-pre/line inputs (together with low-resistance switches), and digitally controlled gain/attenuation ICs for fine-control of input level. We have seen some new takes on re-packaging these functions into one chip, but none are ideal for our particular applications, which require both 0.5dB step control of input gain across the entire range.
We are also using a lot more switching regulators than in the past, preferring them to linear regulators because of efficiency, and also because we have improved methods of filtering any noise generated from such regulators.
John Hanson: Most of the analog components have remained the same. Since all of our new products are digital, we need a very clean analog path, and the traditional components have provided excellent results.
Milo Street: The PGA2500 from TI looks interesting.
2. Are you using any of the modular analog building blocks--mic pres, VCAs and balanced drivers and receivers on a chip, for instance? If so, elaborate where and why.
Marc Lindahl: We're using the SSM balanced receiver/transmitters on Plugzilla, they are just great. A circuit like that depends on temperature tracking--something IC's do very well within the die and package.
I'm looking at a new application for the THAT4320 but I can't really talk about it now!
The other nice building block I used recently is the TLC2932I from Texas Instruments. It's an integrated PLL that performs way better than the usual Varactor circuits.
Michal Jurewicz: Generally no--we belive that our (semi discrete) approach is far more advanced than function chips offered when it comes to the main characteristic of our products: transparent and detailed hi-end sound.
Alan Meyer: We use one specific VCA in our compressor, which is the THAT 2159, but that's nothing new. As stated above, we are considering spending the money for parts like the INA163, but that's still undecided at this time. Given the opportunity there are a few of us here at Alesis that still prefer discrete designs, so modular designs aren't always the most attractive to us.
Tony Agnello: Contemplating the PGA 2500 for a new design.
Don Elwell: We mainly use balanced drivers and receivers. Our analog blocks for those interfaces are proven and thus are pretty cookie cutter now and for us the analog designs pretty much end at the converters.
B.J. Buchalter: No. We have looked at a number of these integrated solutions for mic-pres and differential driver/receivers, but we have not chosen to use them because of either sound quality (e.g. they don't match the "sound" of our discrete solutions), or due to supply rail limitations that make them unsuitable for integration for our purposes.
Nathan O'Neill: The only truly modular chip we use is the Cirrus CS3310, which TI have a pin-compatible cross, the PGA2311. We also use the +/-15V version, the TI PGA2310, which is often a better fit for our high-end designs. Mic-preamps use the TI INA103 or INA162 instrumentation amp, in conjunction with some low-on-resistance switches allowing varying course gain settings.
John Hanson: We've looked into many of the modular analog building blocks, and some have been shown to be very good. Most of the time we are able to achieve as good or better results with our discrete designs and many times the discrete designs are less expensive.
Milo Street: SSM2019/INA217 Mic pre. Few parts, easy to troubleshoot.
3. Are there any newly or recently released components that are improving your analog designs? If so, what are they and in what areas are you seeing improvement (ease of design, performance aspects, etc).
Marc Lindahl: High capacitance ceramic caps are a minor revolution. They have better characteristics than tantalum or aluminum electrolytics and can replace them in all but the highest capacitance or voltage scenarios. They allow circuit layouts to be smaller and impedances to be more tightly controlled, and don't have the nonlinearities of electrolytics.
Michal Jurewicz: There is a lot of new opamps, transistors and some passive components that are superior in performance.
Alan Meyer: Not at the moment, but always on the lookout.
Tony Agnello: Yes. Ease of design, better performance, ease of manufacture, less power, less real estate.
Don Elwell: The continued integration of all the various pieces makes for vastly improved analog designs.
B.J. Buchalter: The Analog Devices ADG452 analog switch has a really low Rds-On and full analog supply rails. As a result, it really enables excellent low noise, low-distortion digitally controlled analog solutions. The performance is key, and it makes designing with the part much easier.
Nathan O'Neill: Switching regulators make point-of-use power supplies a lot easier to design and layout, and offer great improvements in power-supply efficiency. As for analog audio, while the modular analog audio products offer ease of design for novice engineers, in general we haven't seen them as being price-competitive with roll-your-own designs.
John Hanson: No
Milo Street: The above mentioned parts.
4. How would you characterize the differences in making a great analog design versus making a great digital design?
Marc Lindahl: It's much harder to simulate an analog design, so you end up with more trial and error to get it right. The approach is more simulation and modeling oriented when you're working in the digital domain, just because you can, and it's a lot easier. Either way if you don't use your ears, and if you're not willing to work on something until it sounds right, then, well, as they say, "GIGO".
Michal Jurewicz: Depends on approach--digital allows for easier integration of numerous features at no added production cost. However, in our approach, the focus is on sound quality and the approach to design for that is similar in both analog and digital parts of projects. In many ways digital is really a high speed analog. Our ultimate test for both is listening, which is the similar procedure for both domains.
Alan Meyer: It is all the difference in the world. It's the difference between art (analog) and logic (digital). From my experience a good analog design is completed by an engineer with an artistic mind, while a good digital design is completed by a system level designer with superb high-level vision.
In an analog design, the designer is focused on issues like minimized signal path, linearity, noise, slew-rate, gain structure and overall signal transparency. Materials for components, connectors, and wires are of concern. Low-noise power supply performance can be critical.
Conversely, in a digital design, the designer is more focused with issues such as clock jitter, transmission line SI (signal integrity), and edge rate issues for EMC (electro-magnetic compatibility). Today's 32-bit micro controllers require a lot of board space to route the wide busses. Complex digital systems require the designer to strategically place and terminate parts and connectors in such a way that performance is maximized and cost is minimized.
Tony Agnello: Both analog and digital designs require careful layout but analog design can be more finicky. Cross-talk and noise coupling are issues but the primary issue is correctly handling grounding. The designer must minimize digital clock noise injected into the analog signal path.
Don Elwell: Magic! Seriously, the really good analog designers I've known have a knack for being able to use esoteric characteristics or second order effects of devices to their advantage. You don't see that too much in the digital world.
B.J. Buchalter: A great digital design is all about the software. On the hardware side of things, the key to digital design is all about the analog aspects of digital. These are the parts that must be really worried over and engineered. Analog design is all about the game of inches; the key to good analog design for audio is the optimization of the design. The same design with different component ranges can nominally do the same thing, but "sound" very different. With digital, once you have the analog aspects of the digital design nailed down, everything important is a "first-order" effect. With analog, second, third, and fourth-order effects can have a dramatic influence on the sound; it is difficult to make simplifying assumptions --everything is important.
Nathan O'Neill: In analog design, one must be sensitive to gain-staging, distortion, oscillation, and noise-floors, especially in the front-end stage of an ADC design. Also, PCB layout tends to be more critical with analog designs than digital designs, especially when you have both on the same circuit board. However, I think the notion of 'greatness' in design can be accomplished by the best use of technology, and I don't think there's any doubt that digital designs with DSP are the most flexible approach to audio signal processing. Essentially, once you've got a clean signal into your ADC, digital design becomes a question of how good your DSP coding skills are.
Milo Street: Analog requires greater attention to clean power supplies, grounds, and overall layout
5. For digital to/from analog conversion circuitry, it seems to me that the bar continues to be raised incrementally in performance while it becomes easier to make a decent sounding circuit. Circuits lifted right out of application notes can run rings around those found in devices like the very expensive digital tape machines of old. Do you agree that good performance is now easy and inexpensive?
Marc Lindahl: Actually I don't agree that there has been much improvement in the past couple of years. I think we've see more that a certain performance level getting less expensive, rather than a significant performance improvement. Either we've hit a wall (which I doubt) or the economics just aren't there to push things to the last limit. Of course compared to 10 or 15 years ago, sure, it's much much easier to get excellent performance. But "ultimate performance" still takes some work.
Michal Jurewicz: It depends how high the bar is. I'd say in mid-quality analog/digital equipment decent performance can be achieved "by the book" (i.e. easily) and at moderate cost. This is reflected in price of equipment which has plummeted threefold since the mid 90's. However top notch high-end design still requires special selection of parts and a lot of design experience and by no means can be described as easy.
Alan Meyer: Without a doubt, this is true--at least from a performance standpoint. Cost is still about the same for the best performing circuits, but the performance is definitely improved. The performance is largely dictated by the converter performance, but is taken to the limit with optimized surrounding circuits. The converter companies that are dedicated to audio like Wavefront Semiconductor, AKM, and Cirrus have continued to push the envelope in converter technology and make it easier for entry-level companies to produce converters with high-performance.
However, excellent measured performance doesn't always mean excellent sounding performance. Application notes are great, but don't always fill the need sonically. Most of what sounds good in the marketplace can largely be attributed to the designer's focus on sound quality. This goes back to the art of analog design and the endless complexities and tradeoffs that lie within.
Tony Agnello: Absolutely. Reference designs work in large part because there aren't very many things to connect (a higher level of integration means that you can't screw up the connection of the filter to the A-to-D for example). And, you also get lower power requirements and use much less real estate.
Don Elwell: Yes -- see #3 above for why.
B.J. Buchalter: Yes and no. It is certainly easier and less expensive to make very good converters. But to get truly great performance, it is still a non-trivial task to design a great conversion system.
Nathan O'Neill: I would agree that it is certainly easier to achieve higher performance in designs than it was 10 years ago, mainly due to advances in ADC and DAC technology, which are both cleaner and also less-sensitive to layout issues and noise in your system. It is also true that the cost to implement the same performance as 10 years ago has reduced--conversely the market expectation has increased in such a way that our designs still cost about the same to make, but now offer greatly improved performance.
John Hanson: It is easier, although a designer still must understand how the converter works to achieve the best results. Issues such as clock stability, power supply decoupling, and signal routing can affect the design profoundly.
Milo Street: Yes.
5b. What sets superlative converter designs apart from the pack?
Marc Lindahl: Going beyond the "application note" to encompass excellent system-wide design. Considering everything--power supply, I/O, layout, enclosure, etc.--as well as creative aspects like appropriateness to the application.
Michal Jurewicz: The primary factor is the sound quality, because this is the main feature judged by the customer. This, combined with brilliant features at a decent price, is what makes it.
Alan Meyer: This is hard to say as there are many, many factors. For starters, the best designs must have a low-jitter clocking system. Jitter measured from products using an internal clock source is typically low because those products are crystal-based. The more difficult challenge arises when clocking externally. This requires an internal PLL that can recover and form a low-jitter clock. Parts like the Wavefront Semiconductor ADCs and DACs reduce the effects of jitter distortion by incorporating their own internal low-jitter PLL. Aside from clocking, the next area to focus is in the analog input topology. Low-noise, high linearity, and high bandwidth are critical for a transparent sound. A final area of concern is PCB layout, which can make or break your design in terms of noise, distortion, and overall sonic quality.
All the best converters measure well, so the true test that sets one converter apart from another is purely subjective. Listening carefully to the converter's transient response using drums, and it's smoothness with piano and vocals are typical methods for testing if the converter is performing well. All the best converters have been heavily scrutinized by their creators and development teams.
Tony Agnello: Good layout and thorough testing. Careful measurement/testing can/will reveal flaws in the layout.
Nathan O'Neill: Aside from the using the latest and greatest ADC or DAC, attention to layout and jitter reduction in the clock generation circuit are the two most critical items. Proper gain staging is also essential.
Milo Street: Low power consumption, low costs.
5c. What conversion parts are you using and why?
Marc Lindahl: We're using various AKM parts currently. They are excellent performers and very, very cost effective.
Michal Jurewicz: We use converters from different manufacturers depending on their usefulness for a particular design. These are always the highest performance parts from each manufacturer's offering.
Alan Meyer: We predominantly use Wavefront Semiconductor AL1101 (ADC) and AL1201 (DAC) parts. This is because most of our products work at 44.1 KHz / 48.0 KHz. To our ears, no other converter company makes better converters at these sample rates. Within our current line of 96 KHz converter products (ADAT-HD24 and Masterlink), we use the AKM 5393 (ADC) and 4393 (DAC). With regards to the measured specs, these parts are very good but not the absolute best. However, at the time, we chose these parts because we found them to offer some of the best sonic performance for the category. These are now older parts, so we're looking into the latest 192 KHz parts from all companies.
B.J. Buchalter: This is a key aspect of converter design: it is the complete conversion system that is important, not just the converter chip. The same chip can be used in a different system with an incredible difference in the quality of the overall system. The analog that surrounds the converter chip (if the design is based upon an integrated converter chip) is as important, if not more important that the chip itself. Clock management, power supply design and signal conditioning are all critical, and all of these components of the conversion system are analog design problems that exhibit sensitivity to higher order effects. It is the design of these subsystems that is critical to performance, and has a significant impact on the cost of the overall conversion system.
Of course, there is also a significant difference between the different types of conversion parts. Things like dynamic range and distortion performance come at a price. On the D/A side, the integrated converter chip market is driven by the DVD market, and there is such a large volume of parts being consumed by the consumer side of the industry that we are seeing major increases in performance coupled with rapidly decreasing prices. On the A/D side of things, this does not hold true; there is no major consumer demand for super-high quality A/D, so the professional A/D parts still carry a substantial price premium over the "OK" parts. This seems to be especially true for 192k parts; inexpensive 192k compromises the converter quality at all sample rates.
Nathan O'Neill: We use many different ADCs and DACs at the moment across our various production lines. Older products may use the [AKM] AK5392 or [Cirrus Logic] CS5396 ADCs and AK4393 DAC, newer ones the CS5381 ADC and CS4392 (both which offer decreased latency, an increasingly important spec, while offering equal or improved dynamic range and THD performance), while some of our commercial products use slightly more cost-effective parts where price/performance is important (the CS5361 is a good example). Every design is a little different and decisions are made on the product's performance specs as well as the cost of the parts available.
John Hanson: We use a variety of parts from different manufacturers. In our facility, we design products that retail from sub $100 to several thousand dollars. Of course this demands different converters for different applications ranging from low cost CODECs to very high end ADCs and DACs.
Milo Street: Currently Cirrus CS4272 and AKM AK4396. High performance compared to cost and power requirements.
6. There's a new crop of digital I/Os available, and formats such as USB, Ethernet and FireWire are being increasingly used for audio purposes. Are you incorporating new protocols into your designs?
Marc Lindahl: Sure, of course! For example, Plugzilla uses USB for communicating to both the front and real panel subassemblies.
Also, I'm working on some FireWire development boards, which I will make available to other audio companies for experimentation, development, etc.
Michal Jurewicz: Yes, we are- our main focus now is on FireWire.
Alan Meyer: Alesis is only just begun to venture into this realm. We've just introduced a series of affordable mixers called the Multimix FireWire Mixers, which offer multi-channel audio to the computer. We also have recently started shipping some USB audio products including the Multimix USB mixers and the Photon X25, which offer 2 channels of audio to and from the computer over USB 1.1.
Tony Agnello: Yes.
Don Elwell: For USB we use the Cypress EZ-USB and EZ-USB-FX family of devices.
B.J. Buchalter: We were one of the first companies to ship FireWire Audio (almost 4 years ago). All of our digital audio hardware products are built upon our FireWire audio implementation.
Nathan O'Neill: We use CobraNet over Ethernet in our commercial sound products (SymNet) and we are designing a new Lucid product with a FireWire option.
John Hanson: We have incorporated both USB and Ethernet in our designs.
Milo Street: FireWire
6b. What devices are you using for interface?
Marc Lindahl: The AnchorChip USB microprocessor. Very popular because it supports maximum speed USB transfers.
For the Firewire development boards, I'm using the brand new DICE-II chip from Wavefront Semi on one model, and the new Yamaha mLAN chipset on another, so people can experiment with both. I plan to do other chipsets too as time allows.
Michal Jurewicz: We welcome well-functioning chip offerings in the domain of computer interfacing as the R&D required for such custom solutions is prohibitive for small audio companies. There is a lot of activity in the field of audio FireWire for example and we welcome the upcoming DICEII chip from Wavefront Semi as it promises a very powerful and slick solution.
Richard Foss: We are making extensive use of designs with FireWire digital I/O capabilities. The designs incorporate physical and link layer FireWirechips, as well as chips that perform higher level processing of FireWire packets that contain audio samples. The DICE II chip is a good example of such a chip, enabling the flexible encapsulation of audio samples into FireWire packets, and the controlled extraction of these samples. Our task is typically to write firmware that utilizes the capabilities of a chip such as the DICE II, to provide audio and MIDI connectivity between hosting devices. We believe that FireWire, with it's supporting chips, has the potential to radically simplify and enhance the management of connections and control in the field of pro-audio.
Alan Meyer: For the Firewire products, we've given the design win to Wavefront Semiconductor, who distributes a new Firewire (IEEE-1394) IC called the Dice II. The Dice II is an extremely capable and affordable IC for products that require streaming audio over FireWire incorporating a medium-to-high number of channels, high sample rates (up to 192 KHz), with minimized barriers to entry.
For USB, we are using a couple of different ICs made by Texas Instruments (TAS1020B) and Burr-Brown (PCM2902).
Don Elwell: For USB we use the Cypress EZ-USB and EZ-USB-FX family of devices.
B.J. Buchalter: TI Link and PHY, SHARC DSP to handle the Firewire audio layer.
Nathan O'Neill: Cirrus make the CS18101x family of CobraNet chips, and we are currently evaluating several chips for the FireWire design, however our current thinking is that the new Dice II chip (Wavefront) appears to offer both good performance and a great feature set for audio designs.
John Hanson: We are using off the shelf parts.
Milo Street: TI's TSB43CB43 coupled with their TMS320C6713 DSP.
6c. Are they adequate to the task? Any tools you would like to see?
Marc Lindahl: The Anchorchip is more than adequate for what we're doing (UI and display control and MIDI).
For the FireWire stuff, the DICE-II is amazing, it has just about every audio format built into the chip - AES/EBU, ADAT, TDIF.
The DICE-II people had the right idea - they use a GNU tool chain, which means it's all open source, debugged code, and a well known interface, debugger, etc.
In general, I would like to see more vendors support open IDE platforms like Eclipse, for better tool chain integration.
Michal Jurewicz: The more complex the chip function, the more complex the problems and tools. We have no illusion that such an ideal chip with ideal tools exists. This would be very difficult to provide by anyone for such as small market as ours.
Alan Meyer: For now, the Dice II is a very powerful chip that has everything that we could have every asked for. Moving forward, a smaller, less-capable, but lower-cost version would be great for products that don't require a lot of I/O but still want to have FireWire connectivity. The toolset for the Dice II is very extensive, and has filled our needs nicely.
For 2 Channel audio, the USB solutions are adequate. For multi-channel audio, FireWire is a clear winner mostly due to the fact that FireWire has far greater driver support for Windows and OS-X, not to mention the peer-to-peer capabilities.
Don Elwell: We like them--although programming for the 8051 is getting old. I'd love to see them come out with ARM based versions of the devices.
B.J. Buchalter: They have been doing the job for a number of years, and the tools have been up to the task. FireWire audio was a very difficult problem to solve, which is probably why there was only one other company besides us shipping product for a couple of years. Now, there are number of more or less "turnkey" silicon/software solutions, and there has been an explosion of designs and products based upon them--since it is pretty easy to do now. As a result, the transport mechanism has become a commodity; it is the device's capabilities and sound quality (or alternatively, it is the price for simple devices) that are the critical aspects of distinguishing designs with computer based digital I/O.
Nathan O'Neill: We would like to see a gigabit implementation of CobraNet offering more channels and lower latency, as well as a very low cost solution for 2-channel designs. For FireWire, likewise a low-channel count chip would be ideal. The existing tools are quite sufficient for the technology.
John Hanson: They have their limitations, which makes design flexibility difficult, but we are able to work around most of the issues.
Milo Street: Yes.
7. If you are involved in DSP design, why have you chosen the components you use? Do you see advantages in one family of processors over another?
Marc Lindahl: I had used the Motorola 56301 in the Sonorus Studi/o, since it had a built in PCI interface and plenty of horsepower. In the Powercore PCI, I used 56362's because they are cost effective and TC wanted to use the vast 56K DSP code base they had developed for their hardware boxes. Then again, on that board I used an embedded PowerPC because of it's powerful floating point ability. So there, I used a RISC microprocessor for (simple) DSP.
The big advantage for Motorola is the existing base of algorithms already coded. Other than that, anything with floating point is of course much easier to deal with from the algorithm point of view. That's why people are starting to use microprocessors like PowerPC's and Pentium class processors for DSP, like in Plugzilla.
Alan Meyer: DSP plays a major role in our designs. Wavefront Semiconductor DSPs like the 1K (AL3102A) and SCR (AL3201B) play the biggest roles. For 2, 4, or 8-channel products requiring DSP for audio (e.g. EQ, Dynamics, Effects, Reverb, etc.) these parts can't be beat.
For large-scale products, we stick with the major players including Texas Instruments and Analog Devices. For TI, we use the TMS320C6713 and for ADI we use the 21065L (Sharc). The big advantage of TI right now is their highly parallel architecture, coupled with a high-clock rate, memory interface, and floating-point math. The TI really shines for intensive DSP operation.
Tony Agnello: We use Motorola. Originally, it was the DSP of choice for pro audio because its wide date word (24/56 bit) results in wide dynamic range. At this point, we have hundreds of DSP modules written in Motorola code and, to date, that has made sticking with Motorola compelling.
Don Elwell: Because we have vast amounts of code for the one family--that in of itself makes picking the DSP easy.
B.J. Buchalter: Our key considerations have been the ease of integrating the DSP into the types of systems we want to build, and the ease of programming the DSP once it is integrated. We put a lot of effort into evaluating what architecture to work with before we decided on Analog Devices SHARC. Before deciding on the SHARC, we had a lot of experience with the Motorola 56k, and we evaluated a number of alternatives, including processors from TI, but we decided to work with the SHARC for a number of reasons:
1) Programming the SHARC is very easy; it is straight-forward to achieve top performance with the instruction set and execution model.
2) The performance of the SHARC is very predictable, which was critical for estimating the resources required before building hardware.
3) Floating point makes algorithm development substantially easier in most cases. The ability to switch back and forth between fixed point and floating point math on a cycle by cycle basis allows us to maximize the performance of algorithms by choosing the right tool for the job.
4) The SHARC functions effectively as a microcontroller as well as a DSP. This allowed us to integrate the supervisory functions of our device with the signal processing functions. In fact, the entire FireWire control and processing stack is implemented on the SHARC in our products.
5) Performance. The performance oriented aspects of the architecture allow us to implement DSP algorithms that run (in some cases) many times faster than the same algorithm implemented on other architectures.
In addition, although we did not know it at the time we chose the SHARC, Analog Devices has been aggressively pursuing an excellent roadmap for the family, and has delivered (and announced) a number of code-compatible additions to the family that have added significant improvements in performance and functionality. We are very excited about these new parts.
Nathan O'Neill: We are primarily involved in DSP designs, and have made a slow migration from 24-bit fixed point processors to 40-bit floating point processors over the last few years. While many claim that the double-precision mode of 24-bit processors, essentially giving you 48-bit processing and is superior to "32-bit" floating point, our real-world implementations, and tests of similar algorithms on different processors, show its far easier to design with the actual single-precision 40-bit floating point processing available than tying up your processor with double-precision math on a 24-bit fixed point DSP. There are no scaling or normalizing issues to deal with using floating point, and you don't run into digital clipping. Also, the price/performance of floating point is now equal to or better than fixed point, and newer chips offer more flexible audio serial ports as well as features such as DMA, SDRAM controllers, more memory, etc. We use Analog Devices SHARCs in most of our designs (which also offer Link ports for multi-DSP expansion) mainly because of the price/performance and ease of design.
John Hanson: We use different DSPs for different applications. Our low cost designs use our custom ASIC DSP because it is a very powerful, low cost audio DSP. We use off the shelf general purpose DSPs on higher end units for features not available in our custom chip.
Milo Street: TI's TMS320C6713 because of floating point, processing power, and number of audio I/O ports.
7a. Are there advances in DSP technologies that you are particularly excited about?
Marc Lindahl: Most of them are coming from telecom and consumer - cell phones, PDA's, etc. Integrating DSP functions into smaller and cheaper blocks like microcontrollers or even peripherals. On the down side, telecom concentrates on video and low-resolution audio, so most of the best stuff isn't ideal for high-end audio. And much of it is too specialized to use for anything else. But if you're creative you can find ways to adapt it.
Good examples of the application of this type of tech to high end audio are the new DICE-II and DM1500 FireWire chips from Wavefront and Bridgeco, respectively. They both have an ARM core, which is a very fast RISC processor, and they both extend the ARM with DSP instructions, which means you can do things like mixing or dither without using another DSP chip.
Alan Meyer: Most of us at Alesis are always excited to see new DSPs introduced from the major companies like Freescale (Motorola), Texas Instruments, and Analog Devices. The recent lines from TI and some recent introductions from Freescale seem most exciting. However, our greatest excitement comes from future DSPs made by Wavefront Semiconductor.
[An aside question re: the relationship between Wavefront and Alesis: Does Alesis enjoy 'special' benefits due to Alesis and Wavefront belonging to the same family of companies owned by Jack O'Donnell?
Alan Meyer: From an outsider's perspective it would seem that there is a significant price benefit for Alesis by using Wavefront Semi. This is absolutely not the case. Alesis pays the same amount for ICs as all other companies, where price is strictly based on quantities purchased.
Wavefront Semi was originally a division of Alesis. A few years ago, Wavefront was divided off to be more marketable to other audio companies. Wavefront is now its own entity that must show a profit for each company it deals with, including Alesis. The only way this is possible is to charge Alesis the same price for parts as everyone else. Furthermore, most companies like Alesis build in Asia. It is actually our contract manufacturers, and not Alesis, that purchases the parts from Wavefront Semi.
However, there are some real benefits that might be missed by the casual observer. ICs can take years to develop, so Alesis helps guide the direction of new chip development in such a way that Alesis can plan ahead for the next generation of products. Likewise, Wavefront Semiconductor has the inside track as to what chips will be successful without having to spend excess time and money performing market research. Overall, the cooperation alone is what makes our relationship synergistic and both companies independently successful.]
Tony Agnello: Running DSPs on host CPUs (Pentium, PowerPC) is now practical for the majority of audio apps and it's happening everywhere (Plugins, Garage Band, etc). While the notion of running signal processing algorithms 'native' is not new, it has now reached a tipping point.
Don Elwell: You mean besides the low cost!? The dual core ARM/DSP OMAP stuff from TI is kinda cool, but none that are really geared toward pro-audio (yet!?).
B.J. Buchalter: In the SHARC family, the processors have been getting more general purpose compute capabilities along with better and higher performance interfacing. These two improvements are major enabling technologies for us.
Nathan O'Neill: More speed, more memory, more audio channels.
John Hanson: Off the shelf DSPs continue to get more powerful and more affordable. Because of this, future products will be able to achieve better sound at lower prices.
Milo Street: Just the continuing of Moore's Law with increasing performance at lower costs.
8. Any parting thoughts on ICs and their application?
Marc Lindahl: As the part packages get smaller, clock speeds rise, and voltages get lower, the pcb layout becomes more and more critical. It's a very under-appreciated art.
Michal Jurewicz: From the perspective of converter manufacturer I'd welcome an integrated chip providing true state of the art, jitter proof audio clock PLL at a reasonable price. We've been longing for such solution for years.
Alan Meyer: Just like any computer, CAD program, or hammer, ICs are simply a tool that we use to get the job done. Alesis engineering loves the tools that we use, and we love the products that we build. Sadly, end users rarely care what goes on inside the box. Although, in my findings, savvy end-users tend to know what's inside, and take great pride in what they own. For Alesis, we try and use what's right for the product and what will give the users the most amount of enjoyment and will fulfill their needs.
Tony Agnello: ICs have become systems and designing circuits has become systems design.
Nathan O'Neill: ICs have made design easier and allowed engineers to more accurately solve user problems--i.e. better performance, lower cost, more processing per rack-unit, etc. We see the need for faster processors and higher I/O counts in DSPs, and also the need for seamless multi-device expansion to allow for very large matrix mixers (256x256 for example). As long as performance continues to increase, and cost decreases, we think integration of audio functions is generally a positive move.
Milo Street: The complexity of interface protocols such as FireWire and USB requires much more than just silicon. There is quite a bit of software and firmware support that must either be provided by the vendor or developed in-house.
Device Drivers-A Critical Bridge
When interfacing audio gear with computer hardware, the software device drivers can be as critical to performance as the devices they connect. Devendra Parakh, VP of engineering for software developer Singing Electrons, created the first completely full duplex USB 1 driver, worked with Yamaha on mLAN drivers, has done drivers for the BridgeCo 1394 implementation, and is currently leading the driver development effort for the Dice II. He comments on the changes he has seen over the last six to eight years.
From the perspective of device drivers, a lot has changed from USB 1.1 audio to FireWire audio. When audio was introduced with USB, the goal was not professional audio. It was targeted for consumer audio, for streaming applications such as media players, DVD players, gaming and likes of such. And as such, the driver model that Microsoft came up with had little or no concern for things like latencies, synchronization, and external clock sources.
That is where Steinberg's ASIO driver model came into picture and became a de-facto standard for DAW use. ASIO was designed with professional audio applications in mind, and so it provided for determinate (and low) latencies, synchronization, and external clock sources etc... And ASIO was (is!) available on both Windows and Mac(OS9) platforms.
On the Mac side, Apple introduced CoreAudio with OS-X. CoreAudio overcomes most of the limitations of older audio driver models, and is suitable for both consumer and professional audio use. On the Windows side, RTAudio (to be introduced with Longhorn) promises to overcome the current limitations of WDM Audio architecture, and make it useful for professional audio use. Similarly, hardware manufacturers have improved a number of things over time.
With USB 1.1 silicon, there were two big concerns - one was to be able to implement isochronous transfers with low overhead (some applications did it with DMAs, some did it by using fast micro-controllers), the other concern was to avoid clock drift. Most applications used some sort of a PLL or did SRC on the device side. Either technique was expensive in terms of hardware or firmware resources. Very few USB audio devices implemented the technique of slaving the host PC to the device's clock. In fact, most USB audio hardware designers thought (and I've worked with many that still think so!) that it was impossible to do so. We have implemented it successfully in three different devices--one for SMSC and two for Analog Devices.
With FireWire audio, things are quite different. The 61883-6 has been designed from ground up for professional audio use, and it provides for having very accurate presentation times (although the implementation is up to the silicon/device manufacturer).
With the drivers for FireWire audio devices, designers usually have to do more work than they had to for USB audio devices. With USB Audio, it's possible for drivers to simply pass down the audio sample data to the USB driver stack without much manipulation. However, with Firewire audio, the drivers have to repackage the samples, as well as add labels to each and every sample according to the audio format.
At the same time, with proper drivers, it's possible to have multiple FireWire devices connected to the PC and a single application sending and receiving audio from multiple devices, without worrying about clock drift, phase shift, or other timing related issues that crop up when using multiple devices. This opens a lot of possibilities that are not otherwise possible with other kinds of devices.