Overview | Thrust Areas |Development Environment | Case Studies

Thrust Areas

A brief overview of Aricent's Signal Processing Group's thrust areas is listed below:

Domains | Technology

Technology

Aricent possess expertise in the following technologies:

• Multimedia Signal Processing
  
  
• Adaptive Signal Processing

Multimedia Signal Processing

Aricent possesses proven expertise in the following areas:

Speech/Audio Processing

The goal of all speech compression systems is to transmit speech with the highest possible quality using the available channel capacity while maintaining a suitable balance between desired levels of implementation complexity and communication delay. Pulse code modulation (PCM), differential PCM, adaptive differential PCM, delta modulation, adaptive predictive coding (APC) are all examples of waveform coders which strive to reproduce the time waveform of the speech signal as closely as possible.

On the other hand, in the vocoder model speech generation process an excitation signal typical of the air pressure is modulated by the vocal chords and passed through a filter characterizing vocal tract and is transmitted as a set of parameters representing the speech production model. In the receiver, compressed excitation and synthesis parameters are decompressed to reconstruct the excitation vector.

In the gamut of speech coders, Aricent has developed expertise (performance testing and porting on DSPs) on the following:

• TIA/EIA IS-641 ACELP speech codec (at 8 kbps)
  
  
• ITU-T G.729 and G.729A with Annex B CS-ACELP speech codec (at 8 kbps)
  
  
• ITU G.723.1 CELP speech codec (at 5.3 and 6.3 kbps)
  
  
• ITU-T G.726 (ADPCM) speech codec
  
  
• ITU-T G.711 µ-Law/A-Law Compander (at 64 kbps)
  
  
• ETSI GSM speech coders Half-rate(5.6kbps), Full-rate(13kbps), Enhanced Full-rate (12.2 kbps)
  
  
• ETSI /3GPP GSM/3G AMR-NB speech coder (4.75, 5.15,5.9, 6.7, 7.4, 7.95, 10.2 and 12.2 kbps)
  
  
• HNS Proprietary Noise Cancellation Algorithm for IS-641 speech codec
  
  
• HNS Proprietary Frequency Domain Interpolation (FDI) speech codec (at 4 kbps)
  
  
• HNS Proprietary CELP based Low-bit rate speech codec (at 4.8 kbps)

Aricent follows a three-stage strategy in developing speech codecs on DSPs. This includes:

1. The Parallel Simulation strategy: In this strategy, the reference code is run with standard test-vectors and the input/output vectors are tapped for the module(s) of interest. These generated test-vectors are given as input to DSP module(s) and its output is compared with that from the reference code, which has to be same in all cases. After the encoder and decoder modules are checked for compliance with the test vectors, the entire vocoder is checked for bit-exactness with test-vectors.
   
   
2. The Complexity strategy: This strategy is employed to optimize the speech codec for MIPS and memory (data/program) in which the complex/core modules are first considered followed by less complex modules. Some techniques, such as memory over-lay and code/data swapping are also considered in this phase to meet the performance requirements.
   
   
3. The Performance Check: Of the speech codec, where encoder and decoder are run on the evaluation board with different speech inputs of varying durations. This step will also be help in verifying the real-time performance of speech codec in the presence of interrupts and with real-time speech. This can also be used for characterization of speech codec (and can be used to add noise corresponding to different environments)

Image Processing

JPEG corresponds to the ISO/IEC international standard 10918-1, (Digital compression and coding of continuous-tone still images) or to the ITU-T Recommendation T.81. The JPEG standard specifies four modes of operation: lossless compression mode, sequential and progressive DCT-based modes, and multiple-resolution-oriented hierarchical mode. JPEG sets no restrictions on the type of the input colour space. Instead, it views each image as a collection of image components and handles them separately.

Given the human eye’s greater sensitivity to grayscale information than to color, compressing a typical image filters out the colour information, or "chrominance" by a higher proportion than the actual details of shapes, or "luminance". JPEGs therefore, have higher compression ratios for colour images than for grayscale images

JPEG2000 is the new standard to improve the image compression performance. This coding standard is intended to provide rate-distortion and a subjective image quality performance, far superior to existing standards without sacrificing the performance at other points in the rate-distortion spectrum. Lossless and lossy compression, embedded lossy to lossless coding, progressive transmission by pixel accuracy and by resolution, robustness to the presence of bit-errors and region-of-interest coding, are some additional features addressed by JPEG2000. Unlike other standards, this standard uses wavelet transform as oppose to DCT to achieve these features.

Aricent has experience in working on the following Image Codecs ('C' simulation):

• JPEG
  
  
• JPEG-2000
  

Video Processing

To define which fraction of the information content can be eliminated without loss of essential information is a complex task for video streams. Simple forms of redundancy, such as data replication can be encoded efficiently, however, much more redundancy can be jettisoned by an in-depth look at the human perceptive faculty.

The human eye is highly sensitive at low-intensity levels, however, its sensitivity is greatly reduced at high-intensity levels. Using Discrete Cosine Transforms (DCTs), it is easy to separate the high-intensity values from the low-intensity levels because their position in a block (for instance, 8x8) of samples is known. If the transformed coefficients are ordered in the zig-zag sequence, the ordering corresponds roughly with increasing frequency and decreasing visibility.

JPEG, as defined in preceding section, is a compression standard for continuous-tone still images and offers lossless compression mode, sequential and progressive DCT-based modes, and the multiple-resolution-oriented hierarchical mode. Whereas, JPEG treats its picture independently (intra-frame coding), video codecs attempt to predict the current picture from the previous one (inter-frame coding) to exploit the temporal redundancy present in consecutive frames.

H.261, H.263, and H.264 define video coding standards for the compression of the moving picture component of audio-visual services at low bit rates applications, such as videoconferencing/videotelephony. Unlike JPEG, H.261 and H.263 video codecs allows either no random access at all or very restricted random access, thereby increasing the compression ratio. To offer a balance between random access and high compression ratio, MPEG-1 and MPEG-2 video codecs distinguishes among four different coding types, viz. intracoded, predictively-coded, bi-directionally predictively-coded, and DC-coded, for frames. For next-generation coding methods, image compression techniques based on fractals and wavelets are being proposed.

Aricent has experience in working on the following Video Codecs ('C' simulation):

• The ITU-T H.261 Video Codec
  
  
• The ITU-T H.263 Video Codec
  
  
• The ITU-T H.264 Video Codec
  
  
• The ISO MPEG-4 Video Codec
  
  
• The Simulation of Video Switching and Stitching algorithms
  

Channel Coding

Aricent has significant exposure and experience in development and system integration of channel coding algorithms for mobile satellite communications systems and wireless communication systems. The entire channel coding implementations will be developed around Aricent'ss Generic CC Utility Library (GCUL), Generic Channel Coding Library (GCCL), and Generic Service Channel Library (GSCL). All the testing will be done using the Test Harness Library (THALI) that consists of a set of test routines that can independently verify the service and generic channel libraries.

Aricent has different channel coding libraries for 2G/2.5G system and 3G systems. Channel Coding library for 2G/2.5G system (for wireless and satellite) include:

• Convolutional coding and Viterbi decoding Rate={1/2, 1/3, 1/4, 1/5, 3/4) for K={5, 7, 9}
  
  
• Golay (12, 24) encoding and decoding (variant of standard Golay (23,12) encoding)
  
  
• Reed-Solomon encoder and decoder
  
  
• Scrambler / De-scrambler
  
  
• Inter-burst interleaving/de-interleaving and intra-burst interleaving/de-interleaving
  
  
• CRC Generation/Detection
  
  
• BCS precoding/decoding and USF coding/decoding
  
  
• Puncturing and De-puncturing schemes
  
  
• The service library supports all traffic and control channels for 2G/2.5G systems. Ex: for a GPRS based system the service channel library includes
  
  
  • Supports all Packet Data Traffic Channels (PDTCH)
    
    
  • Supports all Packet Control Channels (e.g. PTCCH, PRACH, PBCCH)
    
    
  • Supports GPRS Coding Schemes 1 to 4 i.e. CS-1, CS-2, CS-3 and CS-4
    
    
  • Interface to higher layers

Channel Coding library for 3G system include:

• CRC encoder/decoder
  
  
• Convolutional encoding & Viterbi decoding (1/2 and 1/3 rates), Turbo encoding/decoding
  
  
• Rate Matching/De-Matching
  
  
• First and Second Interleaving/De-interleaving
  
  
• DTX Insertion and Removal
  
  
• Radio Frame Segmentation/De-segmentation
  
  
• Transport Channel Multiplexing/De-multiplexing
  
  
• Physical Channel Mapping/De-mapping
  
  
• Support for Compressed Mode (puncturing, SF/2 and higher-layer scheduling)
  
  
• Service library for 3G User Terminal that supports all the mandatory physical channels

Aricent follows a four-step strategy in developing algorithms for the channel coding library. These include:

1. The Simulation strategy: Whereby the algorithm (transmitter and receiver) is developed as a standalone module in Matlab or 'C' and performance evaluated for different signal-to-noise levels
   
   
2. The Integration strategy: Where all the modules are integrated into the channel coding library and the performance of a channel is evaluated in different signal-to-noise conditions and under various fading scenarios, such as no-fading, AWGN, and Rayleigh/Rician fading .
   
   
3. The DSP Reference Code: Which is then generated from the reference code generated in the above step. This includes developing a fixed-point reference 'C' code and verifying the performance against the reference code
   
   
4. Porting and Optimisation: Which includes porting the DSP reference code into the DSP and optimising it to meet the system MIPS and memory requirements (core receiver algorithms have to be optimised for MIPS).
   

Modems

Digital modulation techniques can be broadly classified as linear and non-linear. In linear modulation techniques, the amplitude of the transmitted signal varies linearly with the modulating signal.

The most popular linear modulation techniques include pulse-shaped Quadrature Phase-Shift Keying (QPSK), Offset-QPSK, and other derivatives of QPSK. Several practical mobile radio communication systems use non-linear modulation methods, such as frequency-shift keying, where the amplitude of the carrier remains constant.

Minimum shift Keying (MSK) is a special type of continuous phase-frequency shift keying wherein the peak frequency deviation indicating the separation between two frequencies is equal to half the bit rate. The MSK spectrum has lower side lobes than QPSK and OQPSK. However, the former's main lobe is wider than that of the latter. In GMSK, the side lobe levels of the spectrum are further reduced by passing the modulating non-return-to-zero data waveform through a pre-modulation Gaussian pulse-shaping filter.

Trellis-Coded Modulation (TCM) is another scheme where coding and modulation is combined and coding gain is derived without increasing the bandwidth required for transmission. For example, V.34 employs similar advanced techniques to select particular sequences of 2D points that simultaneously minimize the transmitted signal power and maximize the probability of correct decoding in the receiver.

Modems can be classified into wireline, that is, modems that are used in landline systems including IP, and wireless, that is, modems that are used in wireless/satellite systems.

In the wireline modems, Aricent has developed expertise on:

• The ITU-T V.21 Channel 2 modem
  
  
• The ITU-T V.27 ter modem
  
  
• The ITU-T V.29 modem
  

In the wireless modems, Aricent has developed a modem library that includes:

• Support for modulation schemes (GMSK / QPSK / Pi/4 CQPSK / Pi/4 DQPSK / OFDM)
  
  
• Baseband filters (such as square-root cosine frequency response) for 2G/2.5G/3G
  
  
• Burst formatting, burst detection, and classification
  
  
• Timing and frequency recovery
  
  
• Channel estimation and equalization (for 2G/2.5G)
  
  
• Channel estimation and rake receiver (for 3G)
  

Aricent follows a three-step strategy to develop wireline/wireless modems. These include:

1. Simulation strategy: Is employed whereby the reference code is simulated in Matlab or 'C' and stand-alone performance of the modem (transmitter and receiver) is measured.
   
   
2. DSP Reference Code: Is then generated from the reference code generated in the above step. This includes developing a fixed-point reference 'C' code using the basic operations and verifying the performance against the reference code (includes selection of size of variables and minimizing errors due to conversion)
   
   
3. Porting and Optimisation: Includes porting the DSP reference code into the DSP and optimising it to meet the system MIPS and memory requirements.
   

- Top-

Adaptive Signal Processing

Equalisers

Aricent has significant expertise in the area of developing equalisers for both wireless and wireline (2G) modems. In a wireline modem, an automatic adaptive equalizer employing the Least Mean Squares (LMS)-based Adaptive and Fractionally-spaced Equalizers algorithm is used in the receiver to equalize the distortion introduced by the channel. Turn-on and turn-off synchronizing signals as per V.27ter / V.29 are used to synchronize the transmitter and the receiver.

Expertise in wireless equalizer includes simulation and development of the Viterbi equaliser for GSM-based systems.

Echo Cancellers

When full-duplex data is transmitted, one of the problems is the undesired feed through of the transmitted data signal into the receiver vis a hybrid present in the subscriber line interface card. A network echo-canceller is an adaptive transversal filter that adaptively learns the hybrid response and generates a replica of that response which is subtracted from the hybrid output to yield an echo-free received signal.

The key measures of performance of an adaptive echo-canceller are speed of adaptation and the accuracy of the cancellation after adaptation. A fundamental trade-off prevails between these two parameters because a longer averaging time is necessary to increase asymptotic accuracy, but slows the convergence rate.

To enhance the performance of an echo-canceller, a non-linear echo suppressor may be used. In addition, a near-end speech detector is accompanied to avoid adaptation of parameters in the presence of a near-end signal. The ITU-T recommendation G.165 addresses the basic requirements of telephone-line echo-cancellers. G.168 provides additional requirements of wider network conditions, such as performance on voice band data, Fax, residual acoustic echo, and mobile networks.

The network echo cancellers that Aricent has significant experience in include the ITU-T G.165/168 compliant echo canceller. The Aricent echo canceller meets the G.165 requirements of convergence, steady state attenuation, double talk detection under low and high level near end speech, and leakage. The product includes NLP and the available echo canceller operates on several tail lengths and operates at moderate complexity.

Aricent's experience in the area of echo cancellers encompasses the following:

• Simulation of components for the echo canceller and generation of test vectors
  
  
• Fixed-point conversion and porting onto DSP meeting
  
  
• Performance testing in real-time systems
  

General Telephony

For user equipment to utilize the services of the general telephone network, it must be able to generate and detect tones that are used for signalling.

At Aricent, considerable work on both generation and detection of tones has been done. In the gamut of general telephony, Aricent has experience on the following:

1. Algorithms for DTMF detector / generator
   
   
2. Algorithms for TONE/MF detector / generator
   
   
3. Country specific tones, such as dial, congestion, and busy
   
   
4. Inband signalling of the DTMF information, based on the speech codec in use
   
   
5. Porting on the customer-specific DSP
   

DSP Libraries

Aricent has implemented a generic and reusable fixed-point math library which can be ported on to any DSP with minimal effort. This library includes:

1. Arithmetic, Log10, Square root, 16 x 32 bit and 32 x 32 bit multipliers
   
   
2. Short division, LaGrange interpolation, Trigonometric functions etc
   
   
3. Generic Viterbi decoder for multiple rates (1/2, 1/3) and various values of K (3,5,7,9)
   

Embedded Firmware and Device Drivers

Apart from the preceding technology domains, the Aricent Signal Processing Group also has enough expertise in developing embedded firmware and device drivers. These are system/processor centric for a given application. Some of these include:

• Control wrappers which take care of task scheduling/partitioning, or use of RTOS , which takes care of task scheduling, memory management et al.c)
  
  
• The control firmware for Inter-DSP communication and communication between the control processor and DSPs (system architecture specific)
  
  
• Boot Loader and dynamic download software (DSP and application specific)
  
  
• Drivers for data exchange among the DSP and associated peripherals (BSP/McBSP, HPI/EHPI, and DMA, etc.)
  
  
• Device drivers for interfacing the DSP with audio codecs and USB etc
  

- Top-

Last updated : March 10, 2006