Shulou(Shulou.com)06/02 Report--

DAVINCI (ARM+DSP) development strategy has not written for more than two years. Since I finished writing DAVINCI DM3730 development strategy, I have little energy to write. I have been busy with DM3730 platform customer customized product design and mass production task.

With the explosive shipments in the first half of 2015-2016-2017, the company had to go all out to support its major customers and ensure the quality of its products. There is also a lovely baby, inside and outside things come under heavy pressure, have to deal with, as a result, blogs and company websites basically stop updating, after all, these days the general environment is industrial misuse the country, speculation "Xingbang", in this case, it is not easy for technology companies to earn some food fees, so there is basically no time to update the blog.

Now that I have hired some employees and outsourced some things to other friends to do, I slowly calm down and continue my DM8127 development work. At the same time, I share some experiences and continue to work on my blog in the development project.

All in all, it's really hard for a person to get through the entire third-party DM8127 IPNC_RDK package designated by TI, because there are too many packages and one's energy is really limited.

When it comes to DM8127 development, our company actually started in 2013. Yes, we bought Taiwan's APPRO DM8127-IPNC (IP network camera) through iRuihe Zhongda. Now iRuihe Zhongda no longer sells this product, because there are relatively few domestic companies developing with this platform, and there is no way. TI's DSP is expensive, and several big companies such as Haikang have relatively large shipments in recent years. The company has been slow to launch DM8127 products for the following reasons:

First, there was not enough money at that time

Second, there is a shortage of talents in this field, which is not easy to recruit.

Third, all the resources of the company fully support major customers to seize a certain market

Fourth, the DM8127 architecture is relatively complex.

Fifth, the 2A-3A algorithm of high-definition CMOS's ISP has a headache, which greatly increases the development cost and time.

Fortunately, at present, I debugged a 2 million IMX SENSOR of SONY, which is 2.8in. The image effect is much better than the previous 720p. I feel very gratified that I also have the motivation to do the development work. These two years of large-scale shipments, a lot of trivial things, I do not know how to come here, the whole person is not in the mood to engage in development work.

DM8127 is a DM81XX series DAVINCI chip with floating-point C674X DSP with relatively low power consumption, with floating-point computing capacity of 4500m FLOPS, specially designed for high-definition cameras. Because of the ISP module, it can process the SENSOR raw image data of RAW DATA as well as DM368-DM385. Ordinary DM8168 and DM8148 do not have this feature.

DM8168 and DM8148 are more suitable for multi-channel D1 intelligent image analysis DVR products, DM8168 and DM8148 power consumption is higher than DM8127. Of course, it is certain that DM8127 consumes more power than the DM3730 we developed before. after all, the strong performance is here, and the corresponding core boards sold on our website have corresponding performance descriptions, and there is no need to describe them here.

The following is only a simple table to get a general idea of the comparison of several commonly used ARM CORTEX-A8+DSP DAVINCI chips. There is no need to compare the 90nm super high-power DM6467T of ARM9+DSP, and there is also only a CORTEX-A8 without DSP DM385/DM388 that is not listed here.

Function

DM8127

DM8168

DM8148/8147

DM3730

DSP

Floating point C674X 750m

Floating point C674X 1G

Floating point C674X 750m

Fixed point C64 + 800m

ARM

A8 + double M3

A8 + double M3

A8 + double M3

A8

Performance

Strong

The strongest (best)

Strong

Lowest

Power consumption

Big

Maximum

Larger

Minimum (excellent)

ISP

Powerful ISP (excellent)

None

None

Weakened ISP

SATA

None

Yes

Yes

None

Network port PHY

Yes

Yes

Yes

None

USB PHY

Yes

Yes

Yes

None

DSP L2

256K

256K

256K

64K+32K SRAM

ARM L2

256K

256K

512K

256K

H264 codec algorithm

Do not take up DSP resources

Do not take up DSP resources

Do not take up DSP resources

Take up a small amount of DSP resources

Built-in face recognition

Yes

None

None

None

DM8127 internal structure diagram:

DM8168 internal structure diagram:

DM8148/DM8147 internal structure diagram:

If you need more detailed functional comparison, you can go to the TI website to download the corresponding DATASHEET materials.

According to reason, DM8127 is a product released by TI in 12 years. The real chip scale mass production began in 13 years, and it should be a relatively backward chip in 17 years now, so why do we insist on doing it?

If you look at other people's mobile phone chips, a new chip that meets the requirements of the flagship machine has been launched in half a year, and it has been updated very quickly. But from our industrial application point of view, TI DSP is different, TI chip vitality is very strong, generally push the chip is more than 10 years of life cycle, such as DM6446 now there are still companies in large-scale shipments, after all, DM6446 is much cheaper than in 2010. TI those TMS320F28XX industrial control chips, used in power systems, motors, power and other control fields, more powerful vitality, several are more than 10 years of life cycle. Not only in industrial applications, but also in the defense industry TI DSP plays a very important role.

The company insists on doing DM8127 because it has been working on the SOC platform for DAVINCI image analysis, and the current market for image recognition (video analysis) is not shrinking, but growing. I have also read the full text of the "New Generation artificial Intelligence Development Plan" released on July 8, 2017, which must be a product of image recognition (machine vision), such as license plate recognition, face recognition is a good product. It brings great convenience to social management.

In terms of video analysis, TI ARM+DSP 's scheme is quite suitable for the transplantation of different kinds of algorithms. It is also suitable for small and medium-sized companies to quickly design products. It still has advantages over the kind of expensive ARM+FPGA. When transplanting off-the-shelf algorithms (such as OpenCV, etc.) to FPGA, you have to find very good people who are very familiar with verilog language and HLS language, and are very familiar with hardware logic. Very familiar with each module of off-the-shelf algorithms. There is also the killer mace floating-point DSPC66X that TI launched in 2014, which uses 32-level pipeline to execute instructions, has been added to the DAVINCI series since 2017, and now launches DM50X, except that ARM is a dual Cortex-M4 super single-chip microcomputer, running TI RTOS (a real-time operating system such as DSP BIOS). This C66X floating-point DSP is super cool for video analysis.

As for how to write the development strategy of DM8127, my cultural level is limited, and my writing skills are even more limited. I can't write according to that professional layout architecture. As there are many things involved in the IPNC_RDK software package, I can only write from scratch.

Such as building Linux HOST development environment, IPNC_RDK overall architecture analysis, BOOT process and application running process and so on. That is, after buying the chip, refer to the development board and design the hardware, how do we develop this platform step by step, and how to make the bare board run automatically?

IPNC_RDK has similar related manuals, but write the user manual is a bit simple, just tell you how to use their IPNC high-definition camera, software package compilation and so on, on how to work the principle of not too much energy to describe, especially their set of McFW, I always feel that the introduction is a bit simple, if not the analysis of the source code, do not know the structure of McFW.

The starting point of the development strategy I want to write is different. If I have time and interest, I will write and share, which may be helpful to the students of graduation design, or to the novice who come out to work. I may be able to meet friends who share a common interest in this respect. see if it is possible to work together to make a good image recognition (video analysis) product. For example, some friends have good image recognition (video analysis) algorithms, software has a good selling point, while hardware design, production, certification, electromagnetic interference, lightning protection, etc. can not be distracted or not professional enough to do, we can provide the corresponding software platform and hardware platform to help customers achieve their selling points and achieve win-win results at the same time. The products we make together can bring benign help to the society, which I think is also a kind of achievement.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.