DSP Libraries
Our team has been developing signal processing libraries for more than 10 years and has excellent expertise in a variety of fields - from radio communications to neuroprocessing. We have worked with all popular commercially available processors, such as TI C55, C64x, C66x; ADI Shark, Blackfin, 21xx; Motorola and others.
We offer services for adapting existing libraries to your hardware and software environment:
- modification for special requirements
- adaptation to an application or a special DSP core
- API migration from one core to another
- reference code, test tools for simulation models run without using processor simulators
- datasets for functional validation and performance verification on your core

Our libraries are usually supplied as part of the IDE (development environment) in source codes and allows you to quickly and easily master a new processor and learn how to write productive software and optimization methods.
Applications
  • audio/speech processing
  • Internet of Things (IoT), robotics
  • communication
  • image processing
  • neuroprocessing
Core DSP
  • HiFi2, HiFi mini, HiFi3/3z, HiFi4, HiFi5
  • ConnXD2
  • Fusion F1, G3, G6, J6
  • Vision P5, P6, Q7
  • Vectra LX
  • BBE 16EP, 32EP, 64EP
  • B10, B20
Data types, performance
  • 8-bit, 16-bit, 32-bit fixed point arithmetic
  • single precision
  • floating point double precision
  • floating point half precision floating point (IEEE-754 16-bit numbers)
  • accuracy of 2 ULP for all floating point types
  • streaming representation formats for improved vectorization
  • datasets up to 10 GBytes for each library for functional validation

The main thing
  • optimization of memory allocation and cache optimization
  • optimal code size
  • specialized set of functions for certain dimensions for optimal performance
  • code conditioning to support different configurations of the processor core
  • all popular matrix decompositions for floating and fixed point processors
  • hundreds of different FFT, including non-base 2 for applications in communication and image processing
Fusion Jx Library
The most versatile DSP core that supports vectorized calculations for all data types from 8 to 64-bit and single/double precision floating point. The best choice for a variety of general purpose applications in automotive electronics, consumer products, the Internet of Things (IoT) and the industrial segment
Main categories:
- Vector/matrix manipulations
- real/complex mathematics
- KIH/BIH filters
- FFT, including 2D and window functions
- Matrix decompositions and inversion
- Statistics
- Smoothing
HiFi5 DSP library
Performance leader for artificial intelligence for audio and speech. Up to 4 times faster in neural processing and up to 2 times faster in audio processing compared to HiFi4. Ideal for digital assistants, information boards and other voice-controlled products.
Main categories:
-KIH/BIH/Kalman filters
-Real/Complex mathematics
-Vector/matrix manipulations
-Matrix decompositions and inversions
-Interpolation/Smoothing -FFT/BCP
(including JPEG accelerator)
-Capstral processing
Vision P5/P6/Q6 DSP Library
Vision cores are specifically designed for complex algorithms of computer vision, artificial intelligence and neuroprocessing. The library accelerates the vectorized calculation of real and complex mathematical functions.
NEW: Functions for FFT with 16/32 bit precision and floating point are included. Our implementations give a gain of 2 or more times, compared to existing implementations from Cadence
B10/B20 DSP Library
The most powerful DSPs signal cores are focused on connected applications such as SDR/Lidar/5G. The library has a wide range of functions that accelerate work with modern communication protocols and neural networks.
Main categories:
- KIH/BIH filters and related functions
- real/complex mathematical functions
- vector/matrix operations
- matrix decompositions/inversions (QR, Cholesky, Gauss-Jordan, LU, SVD, eigenvalues)
- communication (CRC, convolutional encoding, LTE generation/decoding, etc.)
- FFT (including non-base 2)
- Smoothing/interpolation
The library is optimized for the entire range of ConnX family cores - from the most energy-efficient ConnX110 to the powerful B20. All data types are supported - 16/32 bit integer or fixed point and 16/32/64 bit floating point
FusionF1 DSP Library
Tensilica Fusion F1 DSP is one of the most low-consumption cores for widespread use in the Internet of Things IoT and consumer electronics segments. The library contains a large number of functions for robotics, motor control, as well as for speech and audio processing:
Main categories:
-FIR/BIH filters and related functions
-real/complex mathematical functions
-Vector/matrix manipulations
-Matrix appeals
-Quaternions
-Smoothing/interpolation
-FFT/BCP (including JPEG acceleration)
HiFi4 DSP Library
The optimal choice for high-load audio processing systems such as multichannel audio, microphone arrays and echo cancellers, front-ends of digital audio assistants, neural network speech recognizers. Double the performance compared to HiFi3.
The main categories of functions are:
-IIR/IIR filters and related functions
-Real/complex mathematical functions
-Vector/matrix manipulations
-Matrix decompositions and inversions
-Smoothing/interpolation
-FFT/BCP (including JPEG acceleration)
-Capstral treatment
-Image processing
HiFi3/3z DSP library
Energy-efficient DSP for audio applications, super-broadband speech and audio codecs, echo cancellation, neural networks for speech recognition
The main categories of functions are:
-IIR/IIR filters and related functions
-Real/complex mathematical functions
-Vector/matrix manipulations
-Matrix decompositions and inversions
-Smoothing/interpolation
-FFT/BCP (including JPEG acceleration)
Made on
Tilda