- Research Article
- Open Access
Simulation and Emulation of MIMO Wireless Baseband Transceivers
© Pierre Greisen et al. 2010
- Received: 10 June 2009
- Accepted: 25 November 2009
- Published: 10 January 2010
The development of state-of-the-art wireless communication transceivers in semiconductor technology is a challenging process due to complexity and stringent requirements of modern communication standards such as IEEE 802.11n. This tutorial paper describes a complete design, verification, and performance characterization methodology that is tailored to the needs of the development of state-of-the-art wireless baseband transceivers for both research and industrial products. Compared to the methods widely used for the development of communication research testbeds, the described design flow focuses on the evolution of a given system specification to a final ASIC implementation through multiple design representations. The corresponding verification and characterization environment supports rapid floating-point and fixed-point performance characterization and ensures consistency across the entire design process and across all design representations. This framework has been successfully employed for the development and verification of an industrial-grade, fully standard compliant, 4-stream IEEE 802.11n MIMO-OFDM baseband transceiver.
- Orthogonal Frequency Division Multiplex
- Design Representation
- Hardware Description Language
- Functional Verification
- Stimulus File
State-of-the-art wireless communication systems combine multiple-input multiple-output (MIMO) technology with different signaling schemes to increase throughput, link quality, stability, and range . Unfortunately, the introduction of MIMO has also lead to a significant increase in the hardware and system complexity of the baseband signal processing and has multiplied the number of modes of operation supported by state-of-the-art wireless communication standards. MIMO, in conjunction with orthogonal frequency division multiplexing (OFDM), will be employed in existing and upcoming standards such as IEEE 802.11n, IEEE 802.16e , and 3GPP LTE .
1.1. From Research Testbeds to Industrial Grade Prototypes
Such demonstrators and testbeds have been proven to be instrumental research tools in the early phase of the development of new technologies : during this early phase, many factors are still unknown or rely on heavily simplified and yet insufficiently verified theoretical models. For instance, in the MIMO case, the development and standardization of wideband MIMO channel models (e.g., ) has been an iterative process [19, 20]. The ability to avoid model uncertainties in the early days of a new technology is the strength and the justification for the development of research testbeds. Conversely, one of the draw-backs of such research testbeds is the lack of reproducibility. The nondeterministic behavior of the analog part of the design and the lack of control over the noise and the real-world wireless channel realization makes it impossible to fully reproduce or trigger specific test conditions or events. This lack of control makes a comparison with other testbeds or products virtually impossible and is a serious concern for the efficient verification and characterization of industrial products and for research areas that focus on the investigation of complexity-performance tradeoffs.
Thanks to the various testbed research contributions, the knowledge in MIMO communication has reached a level of maturity that allows proceeding to the development of fully integrated transceivers that are compliant to standards that were created for mass products. For the development and optimization of such products, the testbed approach shown in Figure 1(a) must be complemented with a verification environment that delivers fully deterministic and 100% reproducible results and supports a wide range of additional verification objectives that are mandatory for product development. Such an environment is illustrated in Figure 1(b). The basic idea is to separate the endeavor to understand and correctly model the physical environment (using testbeds) from the system development and characterization process.
In a nutshell, the objective of this paper is to describe the design and verification process of a standard compliant baseband transceiver ASIC. In contrast, our previous work in  and  describes a testbed setup which is designed to evaluate the performance of packet-based wireless MIMO-OFDM transmission under real-world conditions. More generally, the major difference compared to other testbeds in the literature (i.e., [5–14, 19, 20]) is the fact that we rely intentionally on statistical channel models and on models for the analog/RF circuitry to allow for better reproducibility and interpretation and comparison of the test results. Also, the complexity of the setup is considerably reduced, since no coping with RF implementation intricacies is required. More on the negative side, the complexity of the software framework of our testbench is higher than for testbeds, but not substantially higher. For instance, the time to build the verification environment described in this paper was roughly a halfman-year, excluding the actual transceiver.
1.2. Related Design and Verification Methodologies and Tools
The DSP design flow employed in this work and reviewed in Section 2.1 involves different design representations, from MATLAB floating-point to register-transfer-level (RTL) hardware description language (HDL) (a similar design flow is described in ). To achieve the best possible hardware efficiency, the refinement from one representation to another was done manually. Nevertheless, some of the concepts in the present work rely on the paradigms from the five-ones approach described in . However, an approach that adheres more strictly to the five-ones paradigm has received increasing levels of attention: high-level synthesis (HLS) (also referred to as behavioral synthesis). The HLS flow uses one high-level language model (e.g., in ANSI C or SystemC) and performs resource allocation, mapping, and scheduling either automatically or semiautomatically, often based on generic architecture templates. A commercial solution for HLS is for example Catapult C by Mentor . Design of wireless communication systems using Catapult C is described in . Beside the incontestable advantages described in , the automatic approach has two draw-backs: first, the quality for complex state-of-the-art designs still cannot be equivalent to the quality obtained by manual optimization by experienced VLSI designers. Second, the dependence on a specific commercial tool is often unwanted for industrial development.
An even higher level approach for describing and verifying complex wireless communication systems starts from virtual prototypes  built from hardware accelerators and DSPs. The physical layer under consideration in this paper is merely a single IP component in such a system which requires a verification strategy by itself. The electronic system level approach is currently not an option for the baseband processing itself since it is operating at the limit of modern process technologies.
This tutorial paper describes the design flow and a verification and characterization framework for a standard-compliant industry-grade IEEE 802.11n transceiver. In particular, we highlight the role of the different design representations and the importance and structure of a corresponding verification methodology. Furthermore, a generic FPGA emulation platform is described, which provides performance characterization through hardware-accelerated Monte Carlo simulations.
The remainder of this paper is structured as follows. Section 2 reviews the VLSI design and verification process used in this work. The different design representations of the transceiver are introduced and motivated. Section 3 describes the verification methodology and the corresponding environment. The focus is on providing a framework that allows for consistent operation of the previously introduced design representations. Section 4 deals with the implementation of the verification framework. In particular, a generic FPGA emulation architecture is presented. Section 5 briefly reviews the IEEE 802.11n reference design and provides implementation figures of the transceiver in the framework and on the FPGA emulation platform.
2.1. Design Flow and Design Representations
The most crucial step in the above described design flow is the floating-to-fixed point conversion, since it has a significant impact on the final performance of the ASIC compared to the theoretical limit. There are several approaches to tackle the fixed-point conversion challenge. One can roughly separate the methodologies into simulation-based approaches and analytical approaches (and a combination of both) . For instance, in  the floating-point model is manually converted into a hybrid code model by defining fixed-point word widths at some "important" locations; the remaining floating-point values are then interpolated by analytical means. A similar approach is described in , but optimized for digital signal processor architectures. The quality of such procedures can be increased by taking into account the target system performance (e.g., the bit-error rate specification) and by iterating until the target performance is achieved . The approach in this work is similar to , but less automated. In essence, word widths at important block interfaces are defined and all other values are deduced. However, we infer the remaining values manually, mainly by MATLAB Monte Carlo simulations or theoretical considerations on the subblock level. Fine-tuning at the system level is performed through Monte Carlo simulations based on the HDL implementation.
2.2. Automated File Generation
Maintaining different design representations is error-prone  when different teams work in parallel on different representations of the same block. To overcome this draw-back, the same person is typically in charge of maintaining all representations of a single block. On the system-level, automatic conversion scripts are employed to maintain important parameters such as data-types, word-lengths, and register address maps in a single database (Figure 2). This measure together with regular simulation runs helps to ensure consistency across different design representations.
2.3. Verification Tasks and Objectives
The development process described above raises several verification objectives.
(1) Performance Characterization: For large systems, such as MIMO wireless transceivers, the performance is not known a priori and can usually not be obtained analytically. Hence, only statistical evaluation methods can be employed and results must be reviewed in comparison to the performance of other candidate algorithms and to solutions known to be optimal. In general, the corresponding characterization process requires a large number of Monte Carlo simulations which are only feasible when simulation runtimes are sufficiently short.
(2) Fixed-Point Accuracy Analysis: Similar to the first objective, performance characterization of fixed-point implementations is realized through Monte Carlo simulations. The results of these simulations are compared to the results of corresponding floating-point simulations to determine the associated implementation loss. Fixed-point design parameters are adjusted according to that loss and simulations are repeated.
(3) Functional Verification: As opposed to the first two objectives, functional verification is concerned with the question whether or not an implementation behaves according to the behavioral specifications of the system. The corresponding evaluation is usually carried out based on a number of predefined testcases for which the expected response of the system is either known a priori or is obtained from a known-good reference model. The latter is often another, more abstract and already sufficiently verified, design representation against which the current design representation is compared. A typical example is the comparison of the HDL model against a software fixed-point model or the comparison against the behavior of the floating-point model or against known expected results under ideal conditions. To achieve sufficient test coverage, functional verification is usually performed with a large number of meaningful but randomly chosen testcases in combination with a set of dedicated and carefully selected testcases .
(4) Regression Testing: For larger projects, regular regression testing must be performed to maintain the integrity of the design and coherence of the different design representations over the entire duration of the project.
Characteristics of different transceiver simulation models.
Far from hardware (HW) (data and control path)
Reference for other models
Golden model for HDL
Same data path as HW
Debug data path
Different control path
Debug control path
Data and control path as in final ASIC
Cycle-accurate: determine latencies
Performance, fixed-point characterization
Accelerated HDL simulation
Large coverage of functional runs
3.1. Verification Flow
(1) Testcase Definition: The starting point for all verification tasks is the creation of a standardized testcase descriptor which defines the operations to be carried out on the transceiver. This description is completely model independent, so that any testcase can be executed on any platform and on all design representations. A simple testcase can for example stimulate the receiver to process a single incoming data packet, while more sophisticated testcases can describe elaborated sequences of interleaved packet transmissions and receptions. Under all circumstances, testcases define all parameters that are necessary for a simulation. These parameters include, among others, the random seeds for the generation of the channel realization through which a packet is transmitted, the specific noise realization, and the transmitted payload data. A key benefit of a complete descriptor is that stimuli generation is fully deterministic and reproducible, which is a prerequisite to ensure consistent and reproducible simulation results and is necessary for debug and regression testing.
In testcases, parameters are defined on a functional level that is independent of a particular implementation of the transceiver. Thus, if a specific feature of the transceiver shall be configured, only a functional abstraction of this setting is provided in the testcase. This is opposed to directly providing values for hardware registers, which need to be programmed in the HDL model but might not exist in other models or may change as the design evolves. The advantage of this abstraction is twofold: testcases are completely model independent and design changes do not affect existing testcases, reducing the effort for testcase maintenance which is particularly important for regression testing.
A typical testcase for performance characterization contains a large set of very similar simulation runs, which differ only in channel realization, noise realization, or received signal strength. To avoid having a different testcase for each of these very similar simulation runs, testcases can contain multiple simulation runs. These runs differ only by few parameters. The missing notion of relative timing among simulation runs calls for an even finer grained structure: each run is again sequenced into phases which can be timed relative to each other. This is required for functional verification of sequences of several transmit and/or receive simulations.
(2) Stimuli Generation: Stimuli generation reads the compact, human-readable testcase definition and expands the testcase into still model-independent stimuli files. To be independent of the details of the implementation, stimuli are described at transaction level. Typical transactions are transceiver configuration, packet transmission, or packet reception. For each such operation, stimuli files provide the information required to operate the design under verification (DUV). The stimuli generation for the testcase structure described in the previous subsection is summarized by the pseudocode snippet in Algorithm 1.
setup testcase data structure
configuration parameters: functional to
10 sequence sub-simulation runs: concatenate
write concatenated sub-simulation runs to file
(3) Testbench Operation: In order to apply the stimuli to the different design representations, an environment must be created that allows stimulating the inputs of the DUV and to record the responses at its outputs. Besides the DUV itself, the environment must also contain all those components that interact with the DUV and must supply signals that are specific to a particular design representation (e.g., clock and reset for an RTL model). The corresponding functionality is provided by a testbench. Ideally, the same testbench would enable the instantiation of all design representations to ensure consistency throughout the entire design process. Unfortunately, this method also requires the use of a single design-language for all design representations. Such a single-language approach has the advantage of allowing for a continuous step-by-step refinement from the original behavioral floating-point model to the final RTL implementation . However, no single language is equally well suited for all design representations. For example, a hardware description language is clearly ill-suited for algorithm development, while a high-level language cannot provide the detailed representation of a parallel architecture that is required to achieve a hardware efficient silicon implementation. Hence, we shall follow an approach that uses multiple design languages as described in Section 2.1. For this approach, consistency of simulations across all design representations is of utmost importance. To achieve this objective, individual, compatible testbenches must be provided as illustrated in Figure 3. By operating different testbenches from the same stimuli files, different models can be compared against each other while ensuring that the same initial conditions apply. This cross verification of design representations is indispensable for the efficient evolution of a design and to ease debugging of more complicated RTL design representations by comparing them against corresponding behavioral golden models .
(4) Consolidation: The consolidation step translates the response files, which may differ in their format depending on the testbench that collected them, into a unified database appropriate for the subsequent analysis. No intelligence or interpretation of the data is provided in this step to ensure robustness against responses from buggy models.
(5) Analysis: The analysis task compares the consolidated simulation result against system specifications or against a reference. Such a reference can be obtained from another simulation run, that is, from the consolidated output of another design representation, for example, for comparing fixed-point against HDL simulations. Alternatively, the reference can consist of expected responses, which are obtained during stimuli generation. For a packet reception simulation for example, expected received data bytes can be compared to the transmitted payload, which is defined in the testcase, and the modulation and coding scheme detected by the receiver can be compared to the one specified in the testcase. The analysis operation itself is defined in a custom way: a database of analysis plug-ins is available, which can be instantiated according to the purpose of the testcase and the verification objective (e.g., calculate and display packet-error rates).
3.2. Testbench Interfaces for an IEEE 802.11n Transceiver
Configuration and control, enabling the configuration of the transceiver and the management of data transmission and reception.Typically, modulation and coding schemes are configurable and need to be controlled by the entity in charge of operating the baseband transceiver. For an RTL model for example, this corresponds to writing (and possibly reading) configuration registers.
Analog frontend control interface, enabling the interaction of the baseband receiver with the analog RF frontend model. Typically, the baseband processor is in charge of controlling variable gain stages in the RF circuitry and, in turn, receives information on the incoming signal strength from the analog frontend.
Data interfaces, for both transmitter and receiver. In transmit mode, payload data is accepted and baseband samples are output to the RF. In receive mode, baseband samples are accepted from the RF, and decoded payload data is output to the system components in charge of handling the higher layer protocols.
Monitors, that enable the observation of internal nodes of the baseband transceiver for the extraction of valuable debug information. While for software models the concept of monitors is quite straightforward, the observation of internal nodes is more involved for hardware design representations which require dedicated infrastructure to provide access to these nodes (see Section 4.3).
In the previous section, the general verification flow was introduced. In this section, the implementation of this flow is described. The focus is on the use of an FPGA emulation platform for accelerated RTL simulations for functional verification and for fixed-point performance evaluation.
4.1. Framework and Simulations
The framework of the verification flow shown in Figure 3 is implemented in MATLAB. The interfaces to the simulation tasks are stimuli files stored to disc during stimuli generation. The interfaces to the consolidation steps are the response files written to disc during simulation.
The floating-point and fixed-point models of the DUV are written in MATLAB and hence can be executed directly from within the verification framework. The HDL model, instead, is written in synthesizable VHDL. Although started from within the framework, the HDL simulation is outsourced to an HDL simulator (Modelsim by Mentor, in our case). The interaction with this HDL simulator simply consists of passing the location of the stimuli files as parameters. In the same way, the HDL testbench is instructed where to dump the response files, so that the MATLAB framework knows which responses to process once the HDL simulation terminates.
FPGA emulation is supported by means of a hardware driver and dedicated low-layer functions (implemented in a foreign language interface provided by MATLAB, called mex functions) that allow stimuli and configuration data to be sent to the FPGA and responses to be collected from the hardware through a PCIe bus. The next section discusses the FPGA testbench in more detail.
4.2. FPGA Emulation Platform
Monte Carlo simulations for performance characterization (e.g., for the extraction of bit-error rates or packet-error rates),
extensive functional verification runs with a large number of dedicated and random tests, to achieve high test coverage, and
regression testing by checking whether new features or bug-fixes affect test results and to ensure coherence between the RTL code and other design representations.
A bus bridge translates the PCIe protocol into the specific protocol supported by the configuration interface of the DUV. The other stimuli are sent from the host PC over the PCIe bus to dedicated stimuli port adapters which apply the data to the input ports of the DUV. Responses from the DUV are collected by response port adapters and are forwarded to the host PC. The number of port adapters is determined by the number of interfaces of the DUV and can be configured at synthesis time. A user-defined portion of the memory space, accessible from the host PC, is assigned to each port adapter. Port adapters are essentially FIFOs with configurable word widths designed to exhibit a handshake data interface for the application of stimuli and for the collection of responses. Type conversion functions translate the bit-vector outputs of the port adapters into arbitrary HDL data types used for the I/O ports of the DUV when applying stimuli, and vice versa when collecting responses. The conversion requires no hardware overhead.
One drawback of the FPGA emulation is the long implementation time for large designs. To alleviate this issue and to make the system scalable to designs with higher complexity (e.g., using better receivers), the DUV can be partitioned over multiple FPGAs. The handshake interfaces between the blocks allow for putting an entire block on a second FPGA while routing the corresponding interfaces via FPGA interconnections. In this way, a block which is currently under construction can be implemented independently of the rest of the design (provided that the interfaces do not change). An automatic partitioning flow can also be set up using Certify , for instance.
4.3. Monitoring of Internal Nodes
The monitoring of internal nodes of the design for debugging and analysis is an important requirement. During MATLAB or HDL simulation, instantiated monitors can simply dump data according to a configuration file that selectively enables or disables monitors. However, for FPGA emulation or even for the final ASIC, providing sufficient visibility into the design is a major challenge. In order to solve this problem, the DUV is equipped with a debug output port, whose bit-width can be configured at synthesis time. Internal nodes connected to monitors can be multiplexed to this port. Which node is observed can be configured using the configuration interface of the transceiver. If a particular node to be observed is wider than the debug port, several addresses are assigned to this node. Each address corresponds to a bit-slice of the node. In this case, multiple simulations must be carried out to reconstruct the complete signal, collecting a different bit-slice during each simulation. The data from the monitors is collected and consolidated together with other responses.
Key figures of the 802.11n design  on different platforms.
The different coding rates, modulation schemes, and number of spatial streams are described by modulation and coding schemes (MCS) and are defined in the IEEE 802.11n standard . The transceiver handles up to four spatial data streams with four antennas both at the transmitter and at the receiver. The design supports a total of 76 MCSs, most of them both in 20 MHz channels (data rates up to 289 Mbit/s) and in 40 MHz channels (data rates up to 600 Mbit/s), with short or long guard interval, in two different packet formats (Greenfield and mixed format), giving rise to hundreds of modes of operation.
The different blocks of the transceiver are arranged linearly and attached to each other by handshake interfaces. This processing paradigm holds not only for the top level blocks shown in Figure 7, but also across all levels of hierarchy. The handshake interfaces allow operating the top-level of the design (as well as the lower hierarchies) at transaction-level. In addition to the welcome side-effect, that plugging in new or revised blocks into the design is much easier with a standardized handshake interface, the operation of different design representations is simplified considerably. In fact, transaction-level operation is equally well suited for timed (e.g., RTL models) and untimed (e.g., MATLAB models) design representations and eases the design of the corresponding testbenches. For instance, transaction-based stimuli alleviate the FPGA emulation, since cycle-accurate delivery of the stimuli is not required which relaxes the requirements on the corresponding testbench implementation.
FPGA resources (total and relative to available).
Port adapters (7 instances)
Compared to HDL simulation, the FPGA platform achieves a speed-up of two to three orders of magnitude. While the clock frequency of the DUV on the FPGA could easily be increased, the main performance bottleneck is due to the file handling operations in MATLAB. The MATLAB fixed-point simulation has a simulation speed that is comparable to the HDL simulation. Compared to FPGA emulation, the behavioral MATLAB floating-point simulation is slightly slower, but on the same order of magnitude. Note that with the proposed monitoring strategy the collection of large amounts of monitor data potentially decreases simulation speed on the FPGA significantly. This is because the number of reserved monitor output pins on the DUV is limited, so that for the observation of wider internal nodes or when monitoring several different internal nodes, the same simulation has to be run several times to collect the required bit-slices. Moreover, activated monitors also decrease the simulation speed of MATLAB simulations, due to additional file handling operations. A typical 802.11n packet reception on the FPGA emulation takes 0.1 to 2 seconds, depending on the packet size and modulation scheme.
Wireless communication testbeds have been instrumental for the investigation of the propagation environment in modern wireless communication systems and for early demonstrations of the technology. However, the testbed approach is often ill-suited for the VLSI development of industrial and standard-compliant products. The aim of this tutorial paper was to describe a design and verification methodology for wireless communication transceivers. The described approach is based on multiple design representations such as floating-point models, bit-accurate behavioral models, and register transfer level descriptions. To ensure consistency across these different design representations, a common verification framework is required. The described framework delivers consistent and 100% reproducible results. For the rapid fixed-point performance characterization, FPGA emulation is integrated into this framework using a generic FPGA emulation platform.
The authors gratefully acknowledge the entire team of Celestrius Inc. Special thanks go to U. Schuster who played a major role in both designing and implementing the framework. Financial support from the Swiss National Science Foundation under project number PP002-119052 is also gratefully acknowledged.
- Foschini GJ, Gans MJ: On limits of wireless communications in a fading environment when using multiple antennas. Wireless Personal Communications 1998, 6(3):311-335. 10.1023/A:1008889222784View ArticleGoogle Scholar
- IEEE draft P802.11n/D2.0 : Wireless LAN MAC and PHY specifications: enhancements for higher throughput. New York, NY, USA, February, 2007Google Scholar
- IEEE 802.16e-2005 : Local and metropolitan networks—part 16: air interface for fixed and mobile broadband wireless access systems, amendment 2: physical and medium access control layers for combined fixed and mobile operation in licensed bands and corrigendum 1. 2006Google Scholar
- E-UTRAN: physical channels and modulation 3GPP Std. TS 36.211, March 2009Google Scholar
- Haene S, Perels D, Burg A: A real-time 4-stream MIMO-OFDM transceiver: system design, FPGA implementation, and characterization. IEEE Journal on Selected Areas in Communications 2008, 26(6):877-889.View ArticleGoogle Scholar
- Murphy P, Lou F, Sabharwal A, Frantz JP: An FPGA based rapid prototyping platform for MIMO systems. Proceedings of the 37th Asilomar Conference on Signals, Systems and Computers (ACSSC '03), November 2003, Pacific Grove, Calif, USA 1: 900-904.Google Scholar
- Wenk M, Luethi P, Koch T, et al.: Hardware platform and implementation of a realtime multi-user MIMO-OFDM testbed. Proceedings of the IEEE International Symposium on Circuits and Systems (ISCAS '09), May 2009 789-792.Google Scholar
- Caban S, Mehlführer C, Langwieser R, Scholtz AL, Rupp M: Vienna MIMO testbed. EURASIP Journal on Applied Signal Processing 2006, 2006:-13.Google Scholar
- Haustein T, Forck A, Gabler H, Jungnickel V, Schiffermuller S: Real-time signal processing for multiantenna systems: algorithms, optimization, and implementation on an experimental test-bed. EURASIP Journal on Applied Signal Processing 2006, 2006:-21.Google Scholar
- Mostafa R, Gozali R, Palat RC, et al.: Design and implementation of a DSP-based MIMO system prototype for real-time demonstration and indoor channel measurements. EURASIP Journal on Applied Signal Processing 2005, 2005(16):2673-2685. 10.1155/ASP.2005.2673MATHView ArticleGoogle Scholar
- Wallace JW, Jeffs BD, Jensen MA: A real-time multiple antenna element testbed for MIMO algorithm development and assessment. Proceedings of the IEEE Antennas and Propagation Society Symposium, June 2004, Monterey, Calif, USA 2: 1716-1719.View ArticleGoogle Scholar
- Rupp M, Mehlführer C, Caban S, Langwieser R, Mayer L, Scholtz A: Testbeds and rapid prototyping in wireless system design. EURASIP Newsletter 2006, 17(3):32-50.Google Scholar
- Ramírez D, Santamaría I, Pérez J, et al.: A flexible testbed for the rapid prototyping of MIMO baseband modules. Proceedings of the 3rd International Symposium on Wireless Communication Systems (ISWCS '06), September 2006, Valencia, Spain 776-780.Google Scholar
- Rao RM, Lang S, Daneshrad B: Field measurements with a 5.25 GHz broadband MIMO-OFDM communication system. IEEE Transactions on Wireless Communications 2007, 6(8):2848-2859.View ArticleGoogle Scholar
- Garcıa-Naya J, González-López M, Castedo L: An overview of MIMO testbed technology. Proceedings of the 4th International Symposium on Image and Video Communications over Fixed and Mobile Networks (ISIVC '08), July 2008, Bilbao, SpainGoogle Scholar
- Kaiser T, Wilzeck A, Berentsen M, Rupp M: Prototyping for MIMO systems-an overview. Proceedings of the 12th European Signal Processing Conference (EUSIPCO '04), September 2004, Vienna, Austria 681-688.Google Scholar
- Rao RM, Zhu W, Lang S, et al.: Multi-antenna testbeds for research and education in wireless communications. IEEE Communications Magazine 2004, 42(12):72-81.View ArticleGoogle Scholar
- Erceg V, et al.: TGn Channel Models. IEEE 802.11 document 03/940r4Google Scholar
- Goud P Jr., Hang R, Truhachev D, Schlegel C: A portable MIMO testbed and selected channel measurements. EURASIP Journal on Applied Signal Processing 2006, 2006:-10.Google Scholar
- Fabregas AI, Guillaud M, Slock D, et al.: A MIMO-OFDM testbed for wireless local area networks. EURASIP Journal on Applied Signal Processing 2006, 2006:-20.Google Scholar
- Weijers J-W, Derudder V, Janssens S, Petre F, Bourdoux A: From MIMO-OFDM algorithms to a real-time wireless prototype: a systematic matlab-to-hardware design flow. EURASIP Journal on Applied Signal Processing 2006, 2006:-12.Google Scholar
- Rupp M, Burg A, Beck E: Rapid prototyping for wireless designs: the five-ones approach. Signal Processing 2003, 83(7):1427-1444. 10.1016/S0165-1684(03)00090-2MATHView ArticleGoogle Scholar
- Graphics M: Catapult C synthesis. http://www.mentor.com/
- Guo Y, McCain D, Cavallaro J, Takach A: Rapid industrial prototyping and SoC design of 3 G/4 G wireless systems using an HLS methodology. EURASIP Journal on Embedded Systems 2006, 2006:-25.Google Scholar
- Belanovic P, Knerr B, Holzer M, Rupp M: A fully automated environment for verification of virtual prototypes. EURASIP Journal on Applied Signal Processing 2006, 2006:-12.Google Scholar
- Kaeslin H: Digital Integrated Circuit Design. Cambridge University Press, Cambridge, UK; 2008.MATHView ArticleGoogle Scholar
- Menard D, Chillet D, Sentieys O: Floating-to-fixed-point conversion for digital signal processors. EURASIP Journal on Applied Signal Processing 2006, 2006:-19.Google Scholar
- Coors M, Keding H, Luthje O, Meyr H: Design and DSP implementation of fixed-point systems. EURASIP Journal on Applied Signal Processing 2002, 2002(9):908-925. 10.1155/S1110865702205065MATHView ArticleGoogle Scholar
- Menard D, Serizel R, Rocher R, Sentieys O: Accuracy constraint determination in fixed-point system design. EURASIP Journal on Embedded Systems 2008, 2008:-12.Google Scholar
- Morales E, Zucchelli G, Barnasconi M, Jugessur N: Novel methodology for functional modeling and simulation of wireless embedded systems. EURASIP Journal on Embedded Systems 2008, 2008:-9.Google Scholar
- Synopsys : Certify. http://www.synopsys.com/Tools/Verification/HardwareAssistedVerification/Confirma/Pages/Certify.aspx
- Burg A, Haene S, Borgmann M, et al.:A 4-stream 802.11n baseband transceiver in 0.13 m CMOS. Proceedings of the IEEE Symposium on VLSI Circuits, June 2009 282-283.Google Scholar
- Synplicity : HAPS-52 Virtex-5 Motherboard. http://www.synopsys.com/Tools/Verification/HardwareAssistedVerification/Documents/haps-52-ds.pdf
- Xilinx : Virtex-5 family overview. http://www.xilinx.com/support/documentation/data_sheets/ds100.pdf
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.