This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

I want to develop a Neural Network for fixed point DSPs

Other Parts Discussed in Thread: TMS320VC5502, TMDX5515EZDSP

Hello,

I am working with low power TMS320C5502. I am a control engineer and implementing Neural Network for controlling and system identification.

Does Texas Instruments support me to develop and standardize a fixed point Fast Neural Network library for real time purposes?

Next, I want to publish that like Imagelib with Code Composer.

Mosayeb Mirani Taradeh

 

  • Hi,

    What level/kind of support are you talking about? Do you have a C5502 hardware to develop your software?

    Regards,

    Hyun

  • I want to prepare a Neural Network library like Neural Network toolbox in MATLAB for fixed-point architecture of Texas Instruments DSPs, In other words, I want to cooperate with you , so I need your financial support by providing this library. Since, I write feedforward MLP with Gradient Descent & RPROP learning algorithms. I need your assistance to see my codes and revise that codes, then everyone be able to use such completed library. Next, I want to develop this library to other neural networks (recurrent networks,RBF,…) fo real-time purposes. I am using Code Composer Studio 3.3. Also, I have a TMS320VC5502 EVM.

    Thank you so much for attention. 

    Mosayeb Mirani Taradeh

  • hi,

    Understood. I have a couple of comments. First of all, you may try with new C55xx board which is C5515eZdsp. It can be found at http://focus.ti.com/docs/toolsw/folders/print/tmdx5515ezdsp.html. TI has been moved to CCS4.xx so that I recommend to use CCS4.x.

    What is your Neural Network library value? What application area can it be?

    Regards,

    Hyun

     

     

  • Hi

    I ordered a TMDX5515eZdsp and it will be in my hand about two or three weeks later.

    This library is included the MLP neural network and it’s online learning algorithms. The feed-forward MLP with no training (it’s weight trained offline) can be used in a large area like Pattern Recognition, Fault Detection and Diagnosis, Nonlinear Control, Image Processing, etc. Everyone can train by his/her own in MATLAB and extract resulted network (weights) and implement them in DSP. There are both online and offline training in MATLAB, but here we just have online training because we use DSP in real-time, then batch training is not acceptable here. This online training can be used for Adaptive Nonlinear Control,  Fault Detection and Diagnosis, System Identification and everywhere the online learning is needed. In other word, we want to implement Neural Network Toolbox of MATLAB in DSP for real-time purposes.

    In the phase one, this library will be included feed-forward MLP with Basic Gradient Descent, Momentum  Gradient Descent and RPROP learning algorithms, also it will be contained Nonlinear arx with online adaptation for nonlinear system identification and nonlinear adaptive control(this kind of MLP have dynamics of inputs and outputs of MLP a kind of recurrent neural network). This phase of library take four months of my time, but I don’t know how much my work value.  I wonder if you could suggest me a reasonable price.

    Finally, E2E is a public webpage, So I prefer that its better to contact with Email?

    My E-mail:mosayeb.m@gmail.com

    Thank you so much.

    Best regards.

    Mosayeb Mirani Taradeh

  • I am still awaiting your reply.
    Thanks.

    Mosayeb Mirani Taradeh

  • Hi,

     

    Thanks for your suggestion but on one in TI is interesting your idea. 

    Regards,

    Hyun

  • Hi,

    I didn't understand your replay. I do my work with CCS4 and send you my results. Next i would appreciate if you let me know your opinion about that.
    Thank you so much,
    Mosayeb Mirani Taradeh