Very high-speed sampling and serial ADCs in embedded systems
Latest-generation active electronically scaled array (AESA) radar systems can have thousands of TX/RX modules. High bandwidth is needed to connect each array element data converter to the FPGAs that process incoming and generate outgoing data streams. New software-defined radio systems use advanced reconfigurable modulation schemes that increase channel bandwidths and deliver unprecedented wireless data rates. To increase the performance of software radio and electronic warfare systems, it has become critical to use efficient, low-power, low-pin-count, FPGA-connected converter interfaces.
As the performances of analog-to-digital converters (ADCs) are quickly improving, the classical approach of transmitting samples through low-voltage differential signaling (LVDS) reaches its limits. LVDS lanes connected to last-generation FPGA I/O are limited to around 1.4 Gbps. As an example, when targeting a four-channel FPGA Mezzanine Card (FMC) with 12-bit analog to digital sampling at 2.6 GSps, a minimum of 96 LVDS lanes is required for data only, without taking into account the clock and service signals. This is not possible on standard VITA 57.1 connectors, and it takes too much I/O resource on the FPGA.
The insatiable demand for data sampling thus has led to the need for the standardization body JEDEC to introduce the JESD204 standard for a high-speed serial link between data converters and logic devices. Serial link data rates have been pushed up to 12.5 Gbps in the last revision “B” of the standard, released in 2011 for higher-bandwidth requirements. This revision also includes provisions for “deterministic latency” of data transfers.
Simplification of design
Compared to the classical parallel approach, the improvements brought by JESD204 are many. By moving from high-pin-count, low-speed to low-pin-count, high-speed serial interface, the overall system design is simplified with a smaller number of trace routes and easier-to-route board designs. The links use 8b/10b encoding, which incorporates an embedded clock, enabling further pin-count reduction by removing the necessity for routing an additional clock line and the associated complexity of aligning an additional clock signal with the transmitted data at high data rates.
In addition, trace-to-trace tolerances are relaxed relative to synchronous sampling parallel LVDS signals. All these simplifications in the design eventually lead to cost reduction. Moreover, this allows reducing the size of the ADC components, as the output pins required for FPGA connection are less numerous using the low differential swing DC-balanced high-speed current mode logic (CML) standard. Thus, it can further increase the number of components on the small FMC footprint.
It must be noted that the last generation of serial ADCs implements digital down-converters with variable decimation ratios that provide filtering and reduce the output data rate. They might include frequency translation stages (numerical controlled oscillators), finite impulse response (FIR) filtering stages, gain stages, and complex-to-real conversion stages. Numerically controlled oscillators (NCOs) and digital mixers allow tuning the center of the bandwidth of interest to baseband. The filtering stages allow filtering the unwanted part of the spectrum. Gain stages allow compensating for mixer and NCO losses. And complex-to-real conversion enables presenting the final real signal of interest.
These down-converter features dramatically reduce the complexity of radio and radar systems, using part of the FPGA resources to implement these functions. The FPGA capacity is fully used for the important signal processing part as beamforming for radars, for instance.
It is important to know the timing relationship between the sampled signal and its digital representation. This timing relationship is affected by the latency of the converter, which is defined for an ADC as the number of clock cycles between the instant of the sampling edge of the input signal until the time that its digital representation is present at the converter’s outputs. This latency is typically in the range of several nanoseconds in classical parallel ADCs. In JESD204B, this latency is increased by the process of serialization even if the speed of the sampling data transmission lanes is much higher. This latency can typically be several tens of nanoseconds.
JESD204B-compliant receivers are outfitted with an elastic buffer that is used to compensate for skew across serializer/deserializer (SERDES) lanes, which simplifies board layout. This elastic buffer stores the data until the data from the slowest lane arrives. It then releases the data from all lanes simultaneously for digital processing. This skew management is possible because the data clock is embedded in the serial data stream.
While the JESD204B standard has simplified multichannel synchronization by using deterministic latency, minimal latency is needed in some applications such as electronic warfare (EW) and radar applications where actions are required immediately after detection. For these applications, the LVDS interface should still be considered, as the JESD204B-compliant data converter’s delay in serializing the data is omitted. However, applications such as radar warning receivers (RWR) or COMINT that are receiver-only applications tolerate the latency brought on by the JESD204B serialization. These applications thus can benefit from the last generations of ADCs driven by the mass market of telecommunication infrastructure, allowing very high-speed sampling and reducing the complexity of the analog part of the system.
FPGA vendors have developed fully compliant JESD204B intellectual property (IP) that can be implemented in their products for communication with the serial ADCs. For example, the JESD204B Xilinx IP supports 256 bytes per frame and 32 frames per multiframe. It can be configured to support up to 32 lanes.
Flexible design follows the fast moving market of ADCs
In combining the technologies available on ADCs including the new JESD204B standard and FPGAs, EW system architects can dramatically improve data sample processing. FMC (VITA 57 standard), promoted by the VITA FMC Marketing Alliance, allows high data throughput and very low latency response between an ADC or a digital-to-analog converter (DAC) FMC and the FPGA, simplification of the design, and above all, the cost-efficient ability to simply retarget an FPGA carrier card design. All that is required is swapping out the FMC module and adjusting the FPGA firmware. That is why the standard has become the open standard mezzanine of choice.
The FMC standard defines a small format mezzanine, similar in width and height to XMCs or PMCs, but around half the length. As real estate is limited, some features have been included in the standard. First, to save space, its primary power is supplied by the FPGA carrier board. During the power-up sequence, the host interrogates the FMC as to what the feeding voltage must be. In addition, FMCs directly connect the I/O devices on the mezzanine to the host FPGA via a high-speed, high-density connector as if the device was on the host itself, leading to logic reduction and saved space.
The first generation of the FMC standard allows up to 160 for high-pin-count (HPC) or 80 for low-pin-count (LPC) “parallel” I/O signals and up to 10 full-duplex high-speed serial connections (along with some clocks). Figures 1 and 2 shown on page 16 depict an Interface Concept ADC FMC, the IC-ADC-FMC, which can be plugged on a Virtex-7 FPGA carrier board, the IC-FEP-VPX3c, featuring eight high-speed transceivers in front of the mezzanine’s high-speed serial (HSS) links.
At the inception of the FMC standard, the HPC specification appeared as satisfactory in terms of the number of allowed I/O. The evolution of ADC technology, as well as the increasingly demanding requirements of EW system designers, has highlighted the need to go beyond this first version of the FMC standard.
An effort to define suitable FMC enhancements is now underway within the VITA 57.4 working group. The focus is on creating a standard with an increased number of HSS links (increased from 10 to 24) operating at increased speed while keeping the existing connector pinout for I/O. The HSS link throughput is targeted at up to 28 Gbps, extending the aggregate bandwidth to the huge level of 672 Gbps, to which the traditional LVDS links can be added. Backward compatibility is ensured by adding to the FMC connector’s outer columns for additional signals without changing the form factor, real estate, or mechanics (mounting holes, thermal interface, and so on). The approval of this new standard is currently in process.
The future of converter digital interfaces
The industry is requiring better performing ADCs, leading to huge sample data flows. The big push of the ADC industry has led to the development of the JESD204B standard. Looking to the future, it is clear that JESD204 is poised to become the industry choice for the digital interface to converters. Each revision has answered the demands for improvements on its implementation and has allowed the standard to evolve to meet new requirements brought on by changes in converter technology. As system designs become more complex and converter performances increase, the JESD204 standard should be able to adapt to meet the new design requirements necessary.