Programmable Audio: Part One Flexible silicon: GUI-programmable audio processors—an EDN BenchPress project
A trend in signal-processing ICs offers simple and direct parametric control and functional programmability. Taking full advantage of flexible chips, however, may demand flexibility in your design methods, as well.
By Joshua Israelsohn, Technical Editor -- EDN, September 29, 2005
| AT A GLANCE |
| The ratings:Flexible silicon parts—great!Support hardware and soft ware—mostly terrific.Documentation—not ready for prime time.Software interfaces are convincing, if occasionally cumbersome.Though the software tools are mostly quite accurate, be sure to confirm with stimulus/response measurements to prevent unwanted surprises. |
|
Sidebars:
Light up the iron |
| What is a BenchPress? This article is the first of a new type of technical feature in which EDN goes beyond the data sheet using instrumented measurements or observations. |
| Click here for Part Two of this project. |
A simple categorization assigns most signal-processing blocks to one of two groups. On the one hand are generic functions, such as op amps and ADCs, which perform one task and may serve in many disparate applications. On the other hand are application-specific blocks, such as GSM transceivers and 802.11 basebands, that have narrow definitions at a much higher level of functional abstraction and, therefore, perform a more complex set of tasks for only one application. An interesting and growing number of signal-processing blocks lie somewhere in the middle. Further scrutiny reveals that the two categories actually lie on a continuum characterized by a dominantly bimodal distribution with a small but decidedly nonzero population between the groups.
The way designers interact with instances from the two large groups is likewise distinct. Tables of parametric performance characterize single-function generic parts under operating conditions that their manufacturers specify. Part of the design task, then, includes deriving expected circuit-level behaviors from topological analysis and components' spec-table performance. On the other end of the spectrum, conformance to industry standards defines many application-specific blocks. Individual parametric measures may be difficult to extract and impossible to control beyond simply choosing one part over another. In the sparsely populated middle, however, lie parts not given to such easy description. These flexible bits of silicon can take on behavioral attributes that depend on an OEM designer's programming or configuration decisions. But, unlike general-purpose programmable devices, these devices operate at a higher level of abstraction than bytes and words, instantaneous voltages, or individual samples. Instead, these devices operate at a level in which the signal is itself parametric, defined by concepts such as, for example, spectral shape or dynamic behavior. In other words, the functions that these parts offer are not fully defined until you decide what you want them to be, and, unlike with general-purpose signal processors, with these devices, you express those decisions in application-relevant parametric terms.
It comes as no surprise that such devices depend on support hardware and software for evaluation, configuration, and programming. In this first BenchPress project, EDN examines the support environments for two devices that exemplify this growing class of signal processor: D2Audio's four-channel, Class D XS-125-4 power-amplifier module and Analog Devices' multichannel, 28-bit AD1940 audio processor. In this study, both devices process audio signals in ways that readily demonstrate these rather abstract concepts and observations. More important, their evaluation or design-support environments reveal that one's design methods may need to vary to accommodate the nature of these components: Though they may arrive with traditional-looking spec sheets, those familiar documents simply don't—and probably can't—tell the whole story. Flexibility has its costs, and one of them is specificity.
The birth of the bench
To get a sense of how these devices and their support environments work in a design setting, I observed two evaluation systems on the bench. The core of EDN's benchtop-test capability for this project is an Audio Precision computer-controlled, dual-domain SYS-2722 audio analyzer—itself the teacher of many good lessons (Figure 1). Audio Precision equipped the analyzer with an optional lowpass filter that conforms to the AES-17-1998 specification (references 1 and 2). The AES-17 filter provides a sharp roll-off for THD+N (total harmonic distortion plus noise) measurements of signal chains that include DACs, signals from which may include a high-level of out-of-band—and thus most often inconsequential—noise (Figure 2). The filter removes this ultrasonic noise that could otherwise interfere with the measurement instrument by overloading gain stages or by tripping autoranging thresholds. We wanted to look at a Class D amplifier, so Audio Precision also provided an AUX-0025 switching-amplifier-measurement filter, which prevents the fast edges present on a Class D output stage from inducing slew-rate limiting in the analyzer's input stages.
Additional bench equipment includes a Wavetek-provided handheld Meterman 24XT DMM (digital multimeter) and a Cove Arts-provided HP-3456 bench DMM, HP-6234A power supply, pair of Tannoy Proto-J monitors, and Technics SL-PG480A CD player. The monitors and CD player provided for signal-path debugging and accompaniment.
The Analog Devices AD1940 evaluation board comes complete with a wall-wart power supply and a USB cable, so with the addition only of common signal-interconnect cables, it's ready to go, right out of the box. Power-amplifier tests, however, require appropriate loads, and, in the case of the D2Audio XS-125-4, the necessary load capacity is sizable: The amplifier can deliver 125W per channel into 8Ω. Commercially available loads that large are hard to come by but simple enough to build (see sidebar "Light up the iron").
The origin of the species
Evaluation boards appear in a number of contexts. IC-design and -product engineers use them to evaluate first-packaged-silicon prototypes. Field-applications engineers use them to help OEM design engineers with part-specific design challenges. Sales engineers use them to demonstrate an IC's capabilities. OEM designers use them as quasiformal development environments. The further along this list you go, the less the operator is intimately familiar with the IC's architecture, modes of operation, programming, strengths, and limitations.
To be sure, the earliest versions of an evaluation board and its software are less polished than those that find their way to customers' hands. The evolutionary track these support tools experience is often divorced from the rigorous product-development methods that the IC enjoys. The evaluation tool's documentation is one area in which this distinction is evident. Though the sophistication of the evaluation tools tends to mimic that of the IC, the sophistication of the documentation tends to follow a different trend—perhaps a trailing indicator of the tool's distribution.
At minimum, any evaluation board should include a map that identifies the location of test points, jumpers, connectors, switches, headers, status LEDs, and any other element of signal, data, power, or human interface. The map, though necessary, is insufficient. The documentation should include a narrative that explains each interface element: settings for switches and jumpers, amplitudes and signaling conventions for test points and connectors, pinouts for headers, and the significance of status indicators—lit, unlit, or flashing. Insufficient documentation results in customer calls to the applications engineer—costly for both the vendor and the customer. This project serves as a case in point: Despite the quality of the evaluation systems, both required application-support calls for questions that the manufacturers could have anticipated and answered in the documentation. Software documentation is no less important. As familiar as GUIs are to anyone in this industry, device- or application-specific elements vary from vendor to vendor and even from product line to product line within a vendor. Functions and attributes that are intuitively obvious to members of an IC-development team may be significantly less so to their customers.
These observations do not in any way mean to beat up on either Analog Devices or D2Audio; they are in good company: It is rare in my experience for any evaluation tool to arrive with the kind of documentation you would expect if the tool were a full-fledged member of the supplier's product line. Good documentation is costly and time-consuming to prepare and maintain. Note, however, that the flip side to the documentation issue is also true: Superior documentation accompanying sophisticated hardware and software helps engineers quickly work their way up a learning curve and reduces the risk to their schedules. They can confirm their understanding of how the IC and its development environment operate. They can spend less time trying to solve unanticipated problems and conceptual disconnects and more time driving their projects to successful conclusions.
Getting connected
Insufficiently documented or not, these signal processors are bound to provide a memorable first experience. My first exposure to a D2Audio Class D amplifier module occurred in 2003 when system architect and D2Audio co-founder Skip Taylor, PhD, brought a prototype unit for EDN to audition. That first amplifier turned in an impressive sonic performance (Reference 3), and several subsequent auditions with various production models have since reinforced the initial impression. I had not appreciated the sophistication of the module's signal-processing capabilities until I had a chance to spend some time with the amplifier on the bench; Canvas, D2Audio's GUI-based programming environment, reveals those capabilities.
A D2Audio Class D amplifier module offers programmable signal-processing blocks in a fixed module-specific topology. The overall structure provides input switching and spectral shaping; a mixer/crossover channel-allocation section; output spectral shaping, delay compensation, low-volume compensation, and dynamic management; channel trims; and a master volume control. Canvas depicts the signal chain in block-diagram form with clickable blocks. Clicking a block launches its associated control panel, and, if the block's function includes spectral shaping, an interactive graph displays a simulated plot corresponding to the current settings (Figure 3).
The mixer/crossover block to a great extent defines the overall function of the amplifier. When you configure it as a 4×4 continuously variable crosspoint matrix mixer, you can route any input channel to any combination of output channels. When you set it as a two-output crossover, the block provides a four-to-two channel selector followed by a two-way active crossover network, appropriate for biamplified-speaker drives. A three-output-crossover mode allows you to select the signal channel—one of four—and provides triamplified-speaker drives. It also mixes the four input channels through independent level controls and delivers a subwoofer output through a tunable lowpass filter. A four-output-crossover mode uses all of the XS125-4's outputs as a quadamplified-speaker drive. The crossover configurations allow you to choose among Bessel, Butterworth, and Linkwitz-Riley filter sections (Reference 4). You can also select Bessel and Butterworth roll-offs of 6, 12, 18, or 24 dB/octave or Linkwitz-Riley roll-offs of 12 or 24 dB/octave.
With signal manipulators on either side of the mixer/crossover block, the XS125-4 architecture appears at first glance to contain design redundancies in both level and spectral controls. Several strategies make different uses of the resources, however, and the appearance of redundancy quickly gives way to an appreciation for having the right controls in the right place along the signal chain. You can invoke the first sets of controls, for example, to correct for deficiencies in the source device or program material and to compensate for room resonances. Two-band tunable tone controls—lowpass and highpass, first-order shelving sections—provide 14 dB of boost or cut. Five-band parametric equalizers provide as much as 6 dB of boost, 30 dB of cut, and Qs of 0.5 to 1442.7. The five bands are identical—each tunable over a 20-Hz to 30-kHz range. The graph that simulates the equalizer's spectral response indicates the interactions of multiple sections. A comparison between the settings and the measured response indicates impressive accuracy in center frequency, cut depth, and Q (Figure 4).
You can apply the controls that follow the mixer/crossover block to compensate for output-device-specific deficiencies, such as driver or cabinet resonances, or placement nonidealities. These controls include three-band parametric equalizers with ranges similar to the five-band blocks, 0- to 3.98-msec delays, dynamics compressors, and level trims. The module rounds delay settings to the nearest multiple of the signal-sampling period. A traditional loudness circuit compensates for changes in human spectral sensitivity that listeners experience at low output levels.
You can manipulate tone-control, equalizer, and crossover sections in real time by adjusting sliders within a control panel, typing values into parameter windows within the control panel, or moving control points on the graph. Equalizers feature band-select and section-bypass switches that allow you to make A-B comparisons between a given setting and the system's flat response.
The evaluation hardware platform communicates with a host PC running the Canvas software through a USB connection. During the software launch, Canvas automatically recognizes the module to which it is connected and presents the block diagram appropriate to that module's architecture. Canvas also operates in a simulation mode without a connection to a module. This feature allows you to develop a module configuration offline. It also allows you to familiarize yourself with the features and capabilities of module types other than those you are already using. If a module connects through the USB link, bench experiments indicate that the software ignores attempts to load files created for other module types.
As well-designed as Canvas is, a number of improvements would enhance the software as a development and operating platform. The most obvious of these improvements derives from the fact that multichannel-audio applications virtually always benefit from the concept of channel pairs. Whether the installation implements simple 2.0 stereo or 7.1 surround sound, the option to manipulate channels in pairs is often an asset. One transparent way of implementing channel pairs would allow you to associate blocks within one channel with like blocks in another channel, so that they track each other through a bilateral control link. This sort of parametric tracking is common in digital-audio consoles to ensure, for example, that equalization settings for stereo pairs always stay synchronized across the pair or to allow precise trimming of a channel pair's gain without disrupting the balance. Such systems also allow for fixed delta settings to accommodate minor mismatches between paired channels. Similar tracking is desirable for channel-level trims and cross-over settings. In lieu of a tracking function, a quick method of copying and pasting the parameter set as a group from one block to another—or, in the case of multichannel blocks such as the mixer/crossover, from one channel to another—would be welcome.
Dynamic-management blocks require a somewhat more sophisticated type of tracking, and, if D2Audio has implemented this feature in its amplifier modules, it isn't evident from either the documentation or the GUI. When a compressor or limiter acts on one channel of a stereo pair, it must communicate the action it takes to the dynamic-management block attached to the other paired channel. Both analog and digital compressors implement this function by means of a stereo-link, or side-chain, port. Lacking this link, a sound source that appears predominantly on one side of the stereo image and that exceeds the compressor's threshold, causes that channel's compressor to reduce the channel gain but leaves the opposite channel unaffected. The result is that the entire stereo image shifts toward the opposite channel until the triggering event has passed, and then it shifts back again. A stereo link and identical compression settings cause identical gain reduction in both channels even if the program material exceeds the threshold on only one of the channels. This behavior maintains the stereo image as it was originally constructed and allows less obtrusive dynamic control.
The choice of crossover-filter type is complex and may well include subtleties unfamiliar to many OEM designers. The Linkwitz-Riley crossover, though increasingly popular, is less broadly understood than Bessel and Butterworth characteristics, which were well-established in the literature long before then-Hewlett-Packard engineers Siegfried Linkwitz and Russ Riley revealed their approach in 1976. In any event, the fact that D2Audio provides the choice of three filter types suggests that the company understands the value of all three. Either a section within the Canvas documentation or a separate white paper explaining D2Audio's view on this topic would help guide OEM designers who use the crossover block.
Despite these few criticisms and suggestions, the Canvas software environment, the D2Audio evaluation hardware platform, and the amplifier modules work together seamlessly and present a shallow learning curve to OEM designers incorporating these amplifiers in their systems.
| References |
|
| Acknowledgements | ||
| Many thanks to Bob Adams and the SigmaDSP team at Analog Devices for their support with the AD1940 evaluation board. Many thanks also to Skip Taylor, PhD, and the folks at D2Audio for their support with the XS-125-4. Most of all, thanks to Bruce Hofer, Alan Miksch, David Matthew, and the Audio Precision design team and configuration groups for the use of the SYS-2722 and for their support of the BenchPress project. | ||
|
Light up the iron |
|
Despite the fact that audio power amplifiers drive complex impedances, their manufacturers often specify their performance under a resistive load. One reason is that the reactive components of speakers' impedances vary from model to model. Additionally, for any given model, the impedance varies across the audio spectrum. No standard impedance exists. Though manufacturers may test power amplifiers with various speaker-simulation impedance models, the amplifier's data sheet must reflect testing under readily replicated conditions, and, for those purposes, the resistive load is generally preferable. There's hardly anything conceptually simpler than a resistive load. But to observe D2Audio's XS-125 under various conditions, I wanted a switchable load that could present 4, 8, or 16Ω to the amplifier. Also, because use of a breakaway is always a good idea when working with devices that can deliver substantial energy, I wanted the load switch to be able to disconnect the load resistors from the amplifier if necessary and to do so faster than I could yank banana plugs out of their respective jacks. I'm unaware of an inexpensive, readily available, commercial load with these characteristics; if you know of one, let me know. So, it soon came time to light up the soldering iron and build one (Figure A). Due to the switching arrangement needed, I switched the load elements with relays. The resistors are all Dale RH-50 8Ω, 50W devices arranged in quads. Because each quad comprises a series set of parallel pairs, the quad's resistance is also 8Ω, but it can dissipate as much as 200W, which gives a moderate margin over the amplifier's 125W/channel capacity. Relay contacts in the figure appear in their de-energized states corresponding to S1 up and S2 in its center-off position. Raising S2 energizes K1 and K3, which connects a pair of quads in series, providing a 16Ω load that can dissipate 400W. Dropping S2 to its lowest position de-energizes K1 and K3 and energizes K2 and K4, which folds the resistor string to form a parallel pair of quads. This arrangement results in a 4Ω load that also can dissipate 400W. S2 is a an Alcoswitch MTL-106E locking toggle switch that requires an operator to pull the toggle bat before it will move to a new position. This requirement prevents an accidental change in the load resistance that the amplifier encounters. Moving S2 to its lower position energizes all four relays, which disconnects the load resistors from the amplifier. Normally, I'd design this type of circuit so that all relays in their de-energized state would disconnect the load from its source. In this case, however, I had limited time for procuring parts and equipment, and I wanted to be sure that I had a working 8Ω load even if the power supply, which I used to produce the relay-coil voltage, arrived late. The availability of the 8Ω, 50W Dale resistors from on-shelf inventory constrained my choice of a parts distributor. The choice and on-shelf inventory of the distributor, in turn, further constrained my choice of a relay. Small, inexpensive PC-mount relays that can switch the load's maximum 10A are not common, so I chose a 12V, DPDT Potter & Brumfield RFE-24012F relay with 8A contacts, and I wired the two poles of each relay in parallel. This gave an acceptable 37% margin. |
| Author Information |
 You can reach Technical Editor Joshua Israelsohn at jisraelsohn@edn.com. |
-
Great article, Joshua. I look forward to the next installment about the AD1940 and Sigma Studio evaluation. I've been looking for a device for similar application. The articles have been very timely.
I think the idea of Bench Press is a very useful one.
Keep up the great work !!
Jim Totten - 2005-15-10 16:25:00 PDT
