September 10, 2015
By Will Strauss
Intel Gets into the Audio DSP Business
From information gleaned from last month’s Intel Developer Forum (IDF), it’s clear that Intel is serious about making DSP chips for speech and audio. Intel’s “Audio Offload Engines (a Hardware DSP)” is touted as providing audio, voice and speech experiences and acoustic innovation…for the PC and tablet market. In addition, Intel offers speaker improvement though “integrated offload engine (a Hardware DSP)” with firmware and software open to third parties.
Intel freely admits that offloading audio functions to specialty DSP chips is the best way to reduce power consumption (compared to performing the same functions on an X86 processor). Intel also introduced the Firmware Development Kit for its Audio Offload Engine.
Although not explained, Intel dropped the hint that the DSP silicon will be based on Tensilica’s Xtensa engine. It also appears that this Audio Offload Engine will be a co-packaged chip with Intel’s next-generation Skylake microarchitecture.
Qualcomm Expands Hexagon DSP offerings
Traditionally, Qualcomm has been the leader in cellphone DSP and in the upcoming Snapdragon 820, three new Hexagon™ DSPs are employed for maximum performance and efficiency. First, the 820’s powerful modem DSP enables the company’s X12 global multimode LTE carrier-aggregation modem capability, initially enabling 3GPP Category 10 operation for 450 Mbps downlink and 100 Mbps uplink speeds.
Through its new Compute DSP Snapdragon 820 introduces new premium tier user experiences including advanced photography, computer vision, VR and machine learning while simultaneously lowering system power requirements by eliminating energy intensive CPU tasks, shifting to the more efficient compute-optimized Hexagon 680 DSP. The 680 features the HVX vector extensions coprocessor for Low Power Compute, Audio, Voice, Advanced Image/Video Processing and Computer Vision. The 680 DSP is a highly efficient multithreaded programmable compute engine that enables concurrent execution of audio and imaging tasks.
An important sensor hub capability is provided through the 820’s Low Power Island, which employs a very low power independent DSP that acts as a sensor hub for positioning, acceleration and more. As a dedicated lower power island with always-on capability drastic power savings accrue over employing the CPU for such functionality.
The Snapdragon 820 begins early sampling through the last half of this year and is slated to ship in consumer devices in the first half of 2016.
iPhone 6s will Keep Apple Ahead
This week, Apple introduced several things, but our main interest is in the new iPhone 6s and iPhone 6s Plus smartphones. Both are powered by Apple’s new A9 3rd generation 64-bit application processor which is said to be 70% faster than the A8 in the “old” iPhone 6. Embedded on the die is the M9 motion co-processor, which is probably an embedded version of the NXP sensor hub/co-processor employed in the earlier iPhone 6. The tech press will be covering most of this in detail but I’ll highlight what else I found interesting.
Cameras have been upgraded, with the main camera upgraded to 12 MP from the 8 MP version in the iPhone 6 and the front-facing camera upgraded to 5 MP. Apple touts the addition of pressure sensing to the touch input, enabling peek into email content or access app features from the home screen icons. Of course, many new photo and video software additions enhance the user experience.
From a communications standpoint, the LTE-Advanced modem supports up to 23 bands and carrier aggregation. Since Qualcomm is the only company that’s shipping that capability, they are likely the modem supplier for the iPhone 6s family. Although not stated, I’ll bet that the LTE-A modem also supports 3GPP category 6 (300 Mbps downlink and 50 Mbps up link). That’s an improvement over the category 4 (150 Mbps downlink and 50 Mbps uplink) modem of the earlier iPhone 6. Now, we’ll have to wait for the teardowns that will soon be available.
Currently, only T-Mobile has generally rolled out a Cat 6 network in the U.S. AT&T and Verizon have equipped only a few select cities with Cat 6 capability. A Cat 6 iPhone capability will no doubt spur the two big guys to broader action.
Balkanization of SoCs
Unless you are a totally integrated company your SoCs have many elements. For example, there’s usually a main CPU from one source (e.g., ARM or MIPS), memories from other sources (like SRAM/DRAM/FLASH), multiple IP cores (typically DSP & MCU), multiple system busses and I/O interfaces (e.g., PCIe, SATA, AMBA, AXI, SCSI, CPRI, MIPI, MOST). A typical SoC may have as many as 100 IP blocks, often purchased from third parties, acquired bundled with EDA licenses, developed in house or re-used from previous designs. Each of these elements typically would like all of the other elements to behave “my way.” Unfortunately, that’s not usually the case and no one person really knows what’s going on inside the entire chip.
Trying to debug these competing “Balkan” states is difficult. But, to do just that, a new company on the block, UltraSoC Technologies Ltd. (Cambridge, U.K.), aims to ensure better behavior among the “warring states” through its novel UltraDebug® modular semiconductor IP platform that allows SoC designers to build an on-chip debug infrastructure tailored to the specific requirements of their system design. It enables holistic “wire-speed” debugging of system software running on chips that incorporate multiple, heterogeneous functional units sourced from many different third party suppliers, as well as custom logic designed in-house.
This week, UltraSoC announced that it has added deadlock detection capabilities within its products for SoC analysis, profiling and debug. The new features allow SoC architects, developers and debug engineers to detect and diagnose hard-to-find corner cases which can cause complex SoCs to hang or stall intermittently and unpredictably, sometimes after days of continuous normal operation. UltraSoC will be demonstrating its new capabilities at ARM TechCon
(10–12 November) and Semisrael Expo (17 November).
China Mobile, Huawei & Qualcomm introduce Uplink Data Compression
Cellphone uplink speeds need to increase, accommodating the growing trend of user-originated data like Instagram and our own videos to YouTube. And up link traffic increases dramatically at major sports and cultural events, as many want to share their experience with others. To enable higher up link speeds, we can use carrier aggregation, and LTE category 6 (Cat 6) for 300 Mbps downlink speeds is becoming mainstream with uplink speeds of 50 Mbps, while upcoming Cat 7 enables twice the uplink speed to 100 Mbps paired with that 300 Mbps downlink speed. But there’s yet more uplink speed to be had.
The industry’s first commercial verification of Uplink Data Compression (UDC), a TDD technology, was demonstrated last month in Hangzhou by China Mobile (the carrier) together with Huawei (the network supplier) and Qualcomm (the modem supplier). The proprietary technology enables operators to maximize LTE-TDD uplink resources, greatly increasing the number of uplink users and achieving significant compression efficiency for various applications. For example UDC-enabled web browsing provided an impressive 70% compression gain.
The LTE modem in the cellphone intelligently compresses the uplink data at the LTE lower layers based on different conditions and application traffic, effectively reducing the amount of data transmitted over the air. On the network side, operators need only to upgrade the eNB software. The UDC technology is fully implemented in the LTE modem and is completely independent from the cellphone’s operating system. Although TDD modems experience the greatest benefit, FDD modems benefit, too. Qualcomm’s next-generation LTE modems are expected to support this feature later this year (likely in the Snapdragon 820).