Addressing Embedded Platform Performance Bottlenecks for USB 3.0

Fractal Realms series. Backdrop of fractal elements, grids and symbols on the subject of education, science and technology

Video-intensive embedded applications such as security and consumer cameras are demanding higher-performance interconnects for faster data transfer. SuperSpeed USB is the natural choice for the next-generation of high-speed embedded platform connectivity.

by Anup Shivakumar, Cypress Semiconductor

As a high-speed serial interconnect, USB 3.0 (SuperSpeed USB) has found widespread usage among personal computers plus peripheral devices such as video monitors and storage peripherals. Embedded designers are finding that highly integrated USB 3.0 devices can address their high-speed internal data transfers. USB 3.0 can also provide a low-cost interconnect to other devices, along with faster data throughput, lower power, smaller design footprint, reduced BOM cost, and shorter development schedules.

High-definition (HD) and ultra high-definition (UHD) video camera designs can illustrate the many benefits of using a standard serial bus such as USB 3.0. Today’s HD/UHD cameras generate real-time data that must be captured, processed locally and then quickly transferred to remote devices such as security monitors and pooled storage services.

Even a 5 Megapixel camera at 24 frames per second (fps) can generate a 2.4 Gbit/s data flow. Designs relying on older serial interfaces such as USB 2.0 and Wi-Fi 802.11n simply cannot keep up with camera data, let alone handle real-time transfers to other devices within the user network.

The lack of high-speed data paths can lead to higher embedded design complexity and cost, such as upsizing frame buffers and requiring larger local storage elements. These design elements can increase overall cost and development times.

USB 3.0 provides an extremely effective choice for today’s high-performance embedded system developers. Its dual-bus architecture allows for communication with legacy devices while SuperSpeed USB can achieve device transfer rates of 5 Gbit/s. In addition, USB 3.0 has 3x more power efficiency than USB 2.0. Due to its broad industry popularity, SuperSpeed USB is a natural choice for both internal and external data path management with the best combination of high performance, reduced power consumption, and cost structure.

Camera system-on-chip (SoC) architectures are highly tuned for gathering and processing data from high-resolution image sensors using specialized on-chip video and image processing DSPs. Design implementations such as those shown in Figure 1 use a secure digital extended capacity (SDXC) interface for local video/image storage, typically in the form of mini/micro SD Cards. HDMI ports provide video transfer to external monitors. The SoC must also manage the on-board LCD screen. USB 2.0 and Wi-Fi ports can provide for external data sharing. Unfortunately, these SoC interfaces cannot keep up with the data rates generated by the image sensors. Custom SoC solutions can lead to long development times due to their design complexity.

cypressfig1

 

Figure 1: Video Camera Architectural Block Diagram

This camera architecture can be highly tuned for writing the processed image or video frame into local Flash memory. Performance and usability are limited by the available SD storage size, coupled with the interface speed by which the content can be moved to a more permanent storage such as a personal computer or external hard drive.

Certain product categories such as commercial security cameras provide limited on-board storage. They instead rely on fast external interfaces such as wired or wireless LAN (Wi-Fi) to transfer their video streams onto a local server. Some consumer cameras provide a removable storage card (mini/micro SD). However, they tend to not be equipped with a local LCD screen for local content viewing. Thus they depend on the recorded content to be transferred to an external device for storage and viewing. These devices often utilize a variety of popular interfaces such as Wi-Fi, USB 2.0, or Bluetooth for data transfer.

Wi-Fi is highly dependent on the availability of a fast reliable network infrastructure. Even an optimized 802.11n network may only be capable of supporting a 300Mbit/s data rate. The actual bandwidth of most 802.11n network installations fall significantly short of this bandwidth due to configuration variations, wireless security, number of channels bonded, etc. This results in an average throughput of 50-80Mbit/s, reaching 100Mbit/s typically only with more expensive commercial-grade equipment.

USB 2.0 has an effective bandwidth of only 280-320 Mbit/s (35-40MByte/s). Even so, most devices can only deliver 200-240Mbit/s. Sustained USB 2.0 transfer rates depend heavily on software driver and platform optimizations.

Table 1 illustrates the expected data transfer time for a 30-minute 1080p HD (H.264) video of ~18GB using USB 2.0 and various Wi-Fi alternatives. Notice that some cases take longer to transfer the video stream than viewing the video stream itself requires!

cypresstable1

 

Table 1: Data Transfer Times for a 30min 1080p HD (H.264) Video

 

Camera OEMs choosing to address this performance bottleneck are left with two choices: (1) Develop their own next-generation ASIC or (2) Wait for next-generation SoCs. Either choice can easily result in an 18 to 24 month product introduction delay. Alternately, designers can upgrade their existing platforms using off-the-shelf SuperSpeed USB components.
SuperSpeed USB can operate 10x faster data than USB 2.0, making it an appropriate serial interconnect to address the video transfer bottleneck. Off-the-shelf controllers such as the Cypress EZ-USB FX3S SuperSpeed USB controller provide on-board ARM9 core, USB 3.0 functionality, plus two storage ports (configured either as SDIO 3.0 or eMMC 4.41) as shown in Figure 2.

cypressfig2

 

Figure 2: Cypress FX3S SuperSpeed USB 3.0 Controller Block Diagram

The FX3S can easily be integrated into a video camera platform by connecting the FX3S to the camera SoC via its programmable GPIF interface (Figure 3). The Cypress FX3S GPIF II interface is a fully configurable parallel interface that connects to external ASICs/SoCs or FPGAs. For additional information on Cypress GPIF II configuration techniques, refer to Application Notes AN75705 and AN75779. In this configuration, the FX3S and its associated SD Cards merely appear as a USB storage device to the SoC. The compressed video data can be stored in the same manner as the current camera implementations without any impact on the software stack.

cypressfig3

Figure 3: Enhanced Video Camera Architecture using Cypress FX3S

This implementation produces several architectural advantages. First, it provides a low-risk, economical means of adding a SuperSpeed USB port. Second, it moves SD Card management away from the SoC, releasing its CPU bandwidth for other critical tasks.

The FX3S GPIF II, configured as a 16-bit multiplexed address/data bus running at 100MHz, provides a ~1.8x data WRITE (Figure 3, Arrow 1) performance improvement over current implementations.

The dual-SD interface on the FX3S can be configured in a RAID 0 configuration using a second mini/micro SD Card, improving performance while providing end users with increased storage. Application Note AN86947 provides performance optimization techniques for USB 3.0 throughput; AN89661 discusses RAID disk design using the Cypress FX3S. The FX3S can deliver a video READ throughput (Figure 3, Arrow 2) of ~720 Mbit/s for a dual SD card configuration. This reduces the transfer rate for the same 18GB video to less than 4 minutes, delivering a significantly higher performance over existing architectures. With the increased size of storage in cameras, longer recording times and transition to higher-resolution video qualities such as 4K, this performance gap will continue to widen significantly.

The addition of the Cypress FX3S controller does not significantly impact battery life. The FX3S is expected to add ~97 mW of active power to the total camera platform power consumption. Assuming a 1160 mAH, 3.8 V camera battery; the FX3S would consume an additional 13.43 mAH (assumes a 95% conversion efficiency) of energy for the same 30 minute video or ~1.2% impact on the battery life. Adding a second SD Card (~270mW) for higher capacity and better performance will impact total battery life by ~4.3%.

As we have shown above, today’s high-performance embedded platforms such as video cameras are facing a performance bottleneck in transferring their content using existing industry strandard interconnects such as USB2.0 and Wi-Fi. This performance gap will cotinue to widen with the continous demand for larger file transfers and higher resolution content such UHD video. SuperSpeed USB is a natural choice for embedded applications that require high-speed device data transfer connections. Developers are left with the choice of waiting for the next generation of SoCs and take the associated schedule and opportunity cost risk or evolve their platforms around commercially available solution to meet the address this platform bottleneck.

Cypress Semiconductor, San Jose, CA. (408) 943-2600. www.cypress.com