Technology Review of TM’s Digital Wireless Audio 1st generation working with eleven engineering Inc
Technology Review of TM’s Digital Wireless Audio 1st generation working with eleven engineering Inc
White Paper: Technology Review of TM’s Digital Wireless Audio 1st generation working with eleven engineering Inc
Introduction
There are a number of design strategies and features that enable TM to deliver digital wireless audio with excellent fidelity, quality of service and coexistence. These include the underlying real-time hardware of the Xinc2™ processor, the TM software (such as the SKAA® operating system for portable applications and the TM2110 operating system for TV/home theater applications), the RF approach and audio data treatment, each of which is reviewed here in some detail.
ISM Bands
Although TM vends products for both the 5.8GHz and 2.4GHz ISM bands, TM is particularly skillful in delivering reliable solutions in the crowded 2.4GHz ISM radio band (2400 MHz to 2483 MHz). This band is very desirable for use since it exists as an unlicensed band in every country in the world (it is the sole unlicensed band that is available worldwide). This band remained unallocated for commercial use for nearly a hundred years because 2.44GHz is a resonant frequency of water molecules (anything containing water will readily accept energy in this band, the water efficiently converting the radio energy to heat). More recently, the 2.4GHz band has come into general use for short range applications in consumer electronics where the path loss problems with humid air, plants and other water-containing obstacles are less of a problem because of the short distances and typically indoor use. Today the band is being crowded by an increasing number of consumer electronic devices including wireless LAN, cordless telephones, security cameras, Bluetooth devices and many others. While the installed base of such devices grows each year, still one of the most significant sources of interference in the band is the microwave oven which uses massive 2.4GHz radio transmissions to heat food. Most microwave ovens flood the band with interference which morphs and moves, affecting nearly every frequency from 2400 to 2483 MHz. The 2.4GHz band is a challenging environment in which to realize a robust-quality real-time audio link. Therefore the use of advanced techniques and protocols is a must.
Narrow Band
Because of the increasing congestion in the 2.4GHz ISM band, the selection of a well-conceived solution strategy for digital wireless audio is key, especially since such a solution must be counted on to deliver an uninterrupted stream of audio in real time. Selection of the wrong strategy can cause either poor Quality of Service (“QoS”), or poor coexistence with other products in the band, or both. One of the keys to TM’s robust QoS and excellent coexistence is its narrow band RF approach. At any instant in time, TM’s footprint in the 83 MHz-wide band is approximately 3 MHz (there is some variation from one TM family to another because different radios are used in different TM products). This footprint is frequency-agile so TM can effectively detect other transmitters in the band and “hop” out of the way. The use of wide-band (10MHz, 20MHz or even wider) solutions for wireless audio is a common trap since such solutions present a large target for interferers, while at the same time contributing significant congestion to the band simply by their own presence. Solutions of this type may work in many home environments today, but are quite likely to fail as the average home fills up with an increasing number of 2.4GHz devices over the next years. One way to “look a few years into the future” of the 2.4GHz band is to operate a candidate solution on a busy electronics trade show floor (as a large show is typically very busy with 2.4GHz traffic). Many companies refuse to demonstrate their 2.4GHz wireless audio solutions in major trade shows since they have poor QoS in a busy 2.4GHz environment. The 2.4GHz ISM band will continue to be an important band for many years to come due to its unique feature of global acceptance. The entire band is only 83 MHz wide, in fact one 802.11 wireless network consumes 1/3 of the band all by itself (or possibly 2/3s for 802.11n). If one or two cordless phones are added, the chances become slim that an empty 10 or 20 MHz remains to support a wide-band audio system. In contrast, an agile, narrow band, adaptively hopping solution can deal with a tremendous amount of congestion including microwave ovens, WLAN, Bluetooth and cordless telephones concurrently. It should also be noted that all families of TM product use radios with very simple modems. Simple FSK modulation is used since more elaborate modulation has proven itself far more susceptible to interference—and interference is nearly always a major factor in the 2.4GHz band. Elaborate modems may be a good choice for general-purpose data networking where the application is not sensitive to big variances in packet transmission latency, but not for a real-time intensive application such as digital audio streaming where latencies must be known, low and unvarying. The value of the above narrow-band, simple modulation approach cannot be over stated. The unique nature of the 2.4GHz ISM band, in that it is available globally, will cause this band to continue to be loaded with an increasing amount of traffic. Wide-band approaches are already showing themselves to be the first to break down under medium to heavy interference conditions. Such heavy design emphasis on QoS performance is certainly warranted given that QoS problems are now known to be the number 1 reason for wireless audio products being returned to retail stores by dissatisfied consumers.
Duplex
TM never simply broadcasts audio data from point a to point b. Although the audio is traveling only in one direction, the RF communication is always bidirectional. The transmit node always maintains a 2-way (duplex) communication with the receive node. This means each receive node can return critical feedback to the transmit node such as if any audio packets arrived damaged (and therefore need to be resent) as well as statistics on RF channel performance. With TM duplex communication is a rule—even in multi-node solutions where there is more than one receive node. The communication between the transmit node and each receive node is always “closed loop”. TM’s operating systems support multiple receivers (in SKAA for example, up to 4)— all of them still maintaining closed-loop communication with the transmit node. Latency and Receiver Sync For an audio delivery system, latency typically matters a lot (for example in Home Theater applications). TM provides fixed latencies which are very stable. In fact sync between 2 receivers in the same system is maintained within 40 micro seconds nominally (output time difference between Rx units in the same system). The absolute latency can be set by adjusting the audio data buffer size. For same-room home theater applications, latencies (Tx to Rx end-to-end) as low as 10 ms are available (easily meeting Dolby’s rule of max 20 ms for home theatre rear speakers). For multi-room audio, latencies may be greater and TM offers up to 40 ms (and the associated QoS boost associated with larger buffers). The key in any true realtime audio system is that latencies are fixed, predictable and exactly repeatable each time the system is powered up.
Protocol
TM has created a particularly strong method of adaptive frequency hopping, for which TM has an issued patent. Frequency hopping sends information on multiple frequencies and works as follows. The Xinc processor keeps track of several A channels and several B channels in the RF domain. A channels are the foundation channels (they are known good—the clearest, bestperforming channels in the band) and so primary information is always sent on them. B channels are experimental channels so redundant information is sent on them—sometimes data gets through and sometimes it doesn’t. The palette of B channels is continually changed, moved from frequency to frequency. Performance stats are kept on all A channels and all B channels. When a B channel is found with better performance stats than the worst of the A’s over a period of time, that B channel is upgraded to A channel status and the worst of the A channels is dropped from the A palette down to the B palette. The WFD protocol has proven its capability in delivering flawless audio even in heavy interference environments such as technology trade show floors. Frequency Hopping is the communication protocol used in all of TM’s operating systems including SKAA and TM2110.
Data Treatment
Typically audio is samples with 24-bit ADCs, with the data path into the Xinc wireless processor being 16 bits x 48 ksamples/sec per full-range channel. Once inside Xinc, the audio is processed with a high-speed data compressor, reducing the audio payload from 768 kbps to 240 kbps per fullrange channel. Note this is a data-reduction step, not to be confused with dynamics compression which is not done by TM at all. It is important to note that this data reduction step is critically important as it enables the narrow-band RF strategy detailed above. The compression is done using an optimized ADPCM-class algorithm developed at TM called HPX™. HPX™ has a THD+N of less than 0.01% across the 20Hz – 20kHz audible range. The HPX codec is employed in all of TM’s full range (20Hz-20kHz) wireless audio solutions including the TM2110 and SKAA® operating systems. The use of HPX to reduce data payloads is key to TM’s high QoS & coexistence strategy—HPX effectively allows the RF footprint to remain very slim and TM’s solutions therefore boast an industry best combination of QoS and coexistence with other devices sharing the band. Note 0.1 audio subwoofer channels are sampled at 6 kHz and are not compressed with HPX (LFE data is sent uncompressed in the TM2110 operating system). The raw data rate of the RF physical devices used in TM’s hardware is 2.0 Mbps. In a 2-channel audio solution such as SKAA, 480 kbps of payload is sent over a 2.0 Mbps physical link. The difference between the physical device data rate and the payload data rate is called the data rate margin (2,000 – 480 = 1,520 kbps). This margin is used to accomplish packet framing, checksums, feedback, control channels as well as transmission of redundant data. The fact that TM’s protocols are designed with a “healthy” data rate margin is one of the key contributors to TM’s well-known robustness. At the receiver, the data, having been received, quality controlled, error corrected and buffered by Xinc is then decompressed and output at 16 bits x 48ksamples/sec per full-range audio channel. So in summary, audio compression is used to achieve a reduced data payload, the reduced data payload enables a narrow-band hopping approach (with simple FSK modem), and the narrow-band hopping approach enables an industry best combination of QoS and coexistence.
Software Approach
TM’s operating systems are built in software on top of the Xinc2 wireless processor hardware. Every operating system is separated into 8 segments, each running on one hardware thread of the Xinc2 wireless processor. Because of Xinc’s architecture, no RTOS (realtime operating system) is required. TM’s operating systems run directly on top of Xinc’s realtime-optimized hardware. Since each of the 8 segments of the operating system run independently from one another, each software segment can be independently optimized without impacting the performance of any other thread. This greatly improves the turnaround time of software iterations and has allowed TM’s protocols and operating systems to reach a state of maturity in a fraction of the time that would otherwise be required. As an example of this, TM has benefitted from over 1,500 software enhancements in its first 2 years of development. The modularity this design approach affords is considerable. For example, an improved audio processing segment can be swapped in with no risk of affecting the RF baseband, protocol or error correction segments. It is as if each software segment was running on its own processor. This thread independence means that TM’s communication protocols and operating systems enjoy a steady stream of upgrades over time, since these can be done in a very controlled and deliberate fashion. To best enable feature/function customization, TM’s operating systems are split into two parts, one called the System Software (which uses 7 processor threads), and the other called the Application (which uses 1 processor thread). Since there is a very clean interface between the two called the API (application programming interface), the customer’s engineer can very easily implement custom functions and features right into the TM modules by modifying the Application code only. The programmer does not have to deal with the complex and timing-sensitive System Software, thereby eliminating the risk of breaking anything in the most sensitive layers of the TM solution. In its TM2110 operating system, TM has recently enabled customers to modify their Application with an easy-to-use scripting language rather than having to code in assembler or C.
Xinc Wireless Processor
TM’s Xinc2 chip is at the heart of every module TM makes. Xinc is a true “wireless processor” with real-time functions, features and architecture specifically engineered for wireless applications (patents pending). Unlike competing technologies, Xinc is a new architecture created specifically for wireless — it is not a conventional serial processor that needs to be adapted for wireless duty with layers of software. It requires no RTOS. Xinc has internal hardware support for interfacing to RF sections (baseband unit) and I/O for handling digital audio bit streams. Xinc has a hardware-multithreaded interrupt-free architecture that’s perfectly suited to wireless applications and other real-time-intensive applications. Xinc delivers 1 MIPS/MHz with absolute consistently. There is no efficiency loss due to pipeline flushing. Xinc’s multithreaded 8-stage pipe never flushes. Xinc’s immunity to loss of pipeline efficiency (which is triggered by conditional branches and interrupt handling in a conventional processor) creates a 2x benefit to the user in terms of useable MIPS (versus conventional serial processors) when running real code. Xinc’s multiple hardware threads (there are 8) enable Xinc to be deployed in real-time applications without the need for an RTOS. This is a money saver and power saver for the customer. The additional MIPS to support the overhead of the RTOS are not required and neither is an RTOS license fee. In terms of MIPS useable in the application, this amounts to approximately another 2x benefit. The multiple threads mean that projects can be split among several engineers/ companies easily. The fact that Xinc is an non-interrupt-based architecture (no interrupts whatsoever) means that Xinc wastes very little time in context switching (pushing registers onto the stack and vice versa). This represents another 2x benefit in useable MIPS. In total, Xinc can deliver up to an 8x (2x2x2) benefit in its payload MIPS in actual wireless applications. Conservatively, a 4x benefit can be counted on. So in a real-time application, a Xinc running at 50 MIPS (at 50 MHz) can typically do the same work as a conventional serial processor running at 200 MIPS. Xinc is also extremely compact — with a core of only 24,000 gates, Xinc breaks MIPS/mW barriers (also approximately 8x benefit vs. conventional) and can practically fit into a hole left by a primitive 8051 8-bit core (20k gates). Xinc achieves this tiny footprint since it’s 8 thread processors are not created through duplication of hardware — they share execution hardware and a pipeline and run out of phase with one another, sharing access to RAM and internal peripherals. Xinc can scale easily. Xinc is a RISC processor with an 8-stage pipe. This means, that raising the bar on Xinc’s maximum MIPS is not a problem in SOCs. Achieving 300-500 MHz operation without elaborate backend tweaking is possible due to Xinc’s innovative real-time architecture. Xinc was designed for speed. That means that Xinc can outpower ARM 7 cores — and Xinc is half the size of an ARM 7 in silicon footprint. Xinc is presently a 16-bit design (data and address) but can be easily migrated to 32 bits. Due to the benefits of multiple threads and interrupt-free design, code can be engineered more deliberately. More deterministic design of code means there’s far less time wasted on integration of code and exhaustive testing for reliability. TM is currently marketing its second generation Xinc2 processor, which has significantly improved upon the first generation Xinc processor by quadrupling its clock speed to 100MHz, adding a basic digital signal processor (“DSP”) that delivers 50 million math-ops per second and incorporating a host of audio-centric peripheral blocks for improved audio I/O – all accomplished in a complete redesign that shrank the die size from 25 mm2 to 9mm2 and reduced its cost by 80%. The Xinc2 is manufactured in a standard 180nm CMOS process. When running all eight threads, its core draws approximately 0.39mA/MHz at 1.8V, with each thread running accounting for approximately 0.023mA/MHz. TM is well along in the development of the successor Cortex M0 processor. To be manufactured using a 90nm low-leakage standard CMOS process, the Cortex M0 will offer substantially improved power efficiency compared with competing solutions; for example, if an 8051 or ARM 7 (running both an RTOS and an application) consumes 80mW to perform a particular function, the Cortex M0 will be able to do the same in only 2 mW (the Xinc2 takes 20mW). More important, Cortex M0 is provisioned with a new audio DSP that significantly outperforms existing solutions, e.g., Analog Devices Sharc, for critical applications such as sample rate conversion, psychoacoustic effects and amplifier control. For the mobile audio applications that are most critical to the TM customer base, Cortex M0 can replace a Sharc processor for one tenth the cost and using one tenth the power. The introduction of powerefficient 32-bit audio DSP capabilities to the Xinc line enables a dramatic increase in integration for mobile applications, further leveraging TM’s audio transport technologies for the coming five years. Recently, TM has been developing a C-language compiler for its Xinc processors and this project is nearing completion.
System In Package
TM ships wireless audio solutions in the format of System in Packages (“SiPs”) which include the Xinc2 processor. A SiP is a tightly-integrated implementation of TM’s wireless audio solution that can be used by customers in the same way as a packaged IC, thereby heightening customers’ ease of designing TM’s products into existing electronic systems and ultimately even improving the manufacturability of the final retail product. Customers find SiPs superior to even single silicon chip wireless audio solutions because they eliminate the need for printed circuit board modules (RF-circuit resistors, capacitors, filters, inductors, baluns, etc. are all included inside the SiP), resulting in a number of benefits including a lower bill of materials cost and the capacity of being added robotically (in contrast to traditional wireless modules must be handled and inserted manually onto a motherboard). TM currently vends 4 models of SiP: TM2100 (2.4GHz with up to 2.1 channels of audio), Saffron (low power version of TM2100), Octavia (low cost, 0.1 audio version of TM2100), and Radish (5.8GHz with up to 2.1 channels of audio). TM is currently developing a 5th SiP, Jasmine, which will be an enhanced version of Radish.
Operating Systems
TM typically uses a single hardware platform for several disparate applications. For example, the TM2100 SiP may be loaded with the TM2110 operating system and sold as a solution for wireless TV speakers or surround sound speakers and subwoofers. Alternately, the same TM2100 SiP may be loaded with the SKAA operating system and be sold as a solution for portable sources such as iPods and smart phones. The very same TM2100 SiP will also be the hardware platform for PAW-700 (future operating system for semi-pro and karaoke microphone and guitar solutions). TM’s numerous firmware operating systems are all developed in house. These operating systems represent a very significant portion of TM’s IP portfolio and are highly optimized for the specific market application they serve. More information on specific operating systems is available in their respective data sheets.
