|
That is why DSP engineer has to rewrite the C/C++ program to remove most floating point operations. This process is known as "porting". All arithmetic operations, all mathematic functions like convolution, square root, logarithms, exponent and many many other must be adopted to use only fixed point data types. This is rather difficult part of the software development process that requires very good mathematic skills and a lot of time. IntegrIT have already spent hundreds of hours optimizing and porting the common mathematic functions and as a result developed NatureDSP libraries that now provide DSP engineers with powerful fixed point tool and allow to save time and project budged. Moreover, we expanded the ibraries with non standard but very useful functions for digital signal processing.
|
 It is commonly known that the most DSPs in consumer devices like mobile phones, PDAs, players are fixed-point as far as fixed point DSPs provide the best solution in terms of power consumption, performance and the price. According to the common practices, DSP engineers start development from system modeling using high-level tools like MATLAB. Then they rewrite the code on C/C++ language where floating point data types used to implement all mathematic operations. Although fixed point DSPs can handle floating point operations itself, but total performance degraded dramatically because fixed-point DSPs optimized for integer data handling only. To get the most from the fixed-point DSP processor the software must not use floating point operations in real-time mode. That is why DSP engineer has to rewrite the C/C++ program to remove most floating point operations. This process is known as "porting". All arithmetic operations, all mathematic functions like convolution, square root, logarithms, exponent and many many other must be adopted to use only fixed point data types. This is rather difficult part of the software development process that requires very good mathematic skills and a lot of time. NatureDSP Signal library now provides DSP engineers powerful fixed point tool that saves development budget and time to market. We already spent hundreds of hours optimizing and porting the common mathematic functions and as a result developed NatureDSP ibraries that now provide DSP engineers with powerful fixed point tool and allow to save time and project budged. Moreover, we expanded the ibraries with non standard but very useful functions for digital signal processing. NatureDSP Signal library is a collection of signal processing routines needed for implementation of typical algorithms for advanced signal processing. Special memory management optimizes performance and allows designer to utilize efficiently cache or DARAM/SARAM spaces and place most memory consuming blocks into the slow memory. Unique "filter-on-the-fly" feature enables fast filter synthesis in a real time. Check NatureDSP Signal datasheet for details. List of supported platforms:
NatureDSP Signal library works bit-exactly on all supported platforms that allows very accurate software simulation. So developer can do the most work on desktop PC comfortably debugging and simulating the software using powerful IDEs like Microsoft Visual Studio. When software is ready the developer simply recompiles the code for the particular target (i.e for TI C55) and links it with appropriate NatureDSP library. That is all! The software works bit-exactly with the PC version! Note that the hardware is needed only on the final development stage for testing and MIPS measurements. This approach makes software development process much more predictable and manageable. With NatureDSP library there is no need to have expensive hardware for software development. |
IntegrIT
Nature DSP Signal+ for Tensilica HiFi2 is a
collection of signal processing functions needed for implementation
of typical digital signal processing algorithms. All functions are
manually optimized for Tensilica HiFi2 DSP utilizing full data bus
bandwidth and computing power of the HiFi2. With dramatic penetration of multimedia services in our life consumer devices have to meet growing requirement to signal processing capabilities. As a result CPU must effectively manage not only general tasks (GUI, networking, PIM applications etc.) but much more computationally effective tasks as audio/video compression and decompression, speech encoding, acoustic echo cancellation, noise suppression and more and more. To be successful on the market software developers have to become more DSP oriented and hard real time aware. To facilitate software development of hard real time DSP applications IntegrIT has developed a special port of its well known cross platform NatureDSP Signal+ library for market successful Tensilica HiFi2 DSP. IntegrIT Nature DSP Signal+ is a collection of signal processing functions needed for implementation of typical digital signal processing functions which effectively utilize 24bits HiFi2 DSP nature. It contains plenty of highly optimized routines for filtering, FFT, matrix, trigonomentic etc. operations. With IntegrIT NatureDSP Signal+ library software developer don’t need to understand details of algorithms or details of DSP architecture. The library comes with bit exact x86 version which allows rapid software development and simulation in developer's favourite IDE. This approach dramatically shortens product time to market. IntegrIT NatureDSP Signal+ library is already available for ConnX D2, HiFi2 and BBE16 cores. |
||||
IntegrIT
Nature DSP Signal+ for Tensilica ConnX D2 is a
collection of signal processing functions needed for implementation
of typical digital signal processing algorithms. All functions are
manually optimized for Tensilica ConnX D2 DSP utilizing full data bus
bandwidth and computing power of the ConnX D2. It outperforms TI's
C55xx DSPLIB by 45-65% for FFTs and provides up to 4x acceleration
for vector mathematics. The important feature is availability of
bit exact library for x86 PC which makes development and simulation
of complex DSP algorithms simpler and much more straight forward. |
||||
