Posted on

netpp node evaluation platform

netpp node [Spartan6-LX9]

I am glad to announce a new user evaluation platform module called ‘netpp node’. Its motto is ‘IoT on FPGA done right’. See detailed specs and preliminary order information here: [ refdesigns/netpp_node ].

Update [21.12.]:

The netpp node engineering samples v0.0 have just passed the long term burn-in. Running since approx. 11 weeks non-stop, the units are flooded with netpp requests from an embedded PC and have shown no failure in the hardware, except a reboot resulting from a power outage.

v0.1 series [18.1]:

Received the series! So finally we can ramp up with the delivery to beta developers…(thanks for being so patient).

Analog I/O

ADC10 low level control

For analog I/O, U3 on the board is by default populated with a MSP430G2553, functioning as a smart ADC that is controlled from the ZPUng ‘dagobert’ SoC via i2c. All relevant ADC configuration registers are directly accessible via netpp. For instance, we access the low level registers through a process browser panel as shown above to play with the parameters. The process view panel automatically updates the volatile properties from the netpp peer device. The ADC10 variant of the netpp node provides up to six¬† analog channels internally sampled at up to 200ksps. When in synchronous acquisition configuration (SPI master), only five channels can be used.

Differential 16 bit sigma-delta ADC

SD16 analog input

The alternate population option with a MSP430F2013 provides a Sigma-Delta 16 bit ADC with differential inputs and programmable gain amplifier. This variant provides three different input channel configurations using the provided analog input pins on this board. Moreover, the internal temperature is available in a separate channel.

‘Push on demand’ data streaming

By default, the analog sensors are polled, i.e. a measurement value is delivered upon request by the master. For synchronous sampling however, a ‘push’ strategy might be desired, where a netpp node delivers a value stream to a data logger or database. This can be netpp (where the netpp node acts as a master), however for high speed data transfers (‘network scope’), a low overhead UDP stream is more desirable. The dagobert SoC features a data port option with programmable slots to stream I/O channels as well as analog values using a standard real time protocol with 90 kHz time stamps.

Monitoring netpp packet performance

Packet behaviour in a real network is measured using the Wireshark protocol analyzer.

The figure below shows some example netpp transaction log that the netpp node handles at a very low CPU overhead based on direct register accesses.The red bars is the effective number of query responses using somewhat ineffective ping-pong requests. The performance can be increased by accumulating data into larger buffer properties.

For i2c or SPI transactions however, the packet rate is expected way lower.

For high speed performance like MJPEG video streaming, a separate UDP/RTP queue can be set up within the firmware to reach maximum throughput. However, there is no handshaking using this method.

The image below shows a repeated property query from within Python. The pauses are introduced by external disturbance (stress test) that causes a packet drop – and the netpp engine to timeout and re-synchronize.

Python property query session
Python property query session

Improved RX/TX queue

With an improved packet FIFO on FPGA, I was able to crank up the number of netpp requests per second, as shown in the Wireshark trace below. This test makes sure that several netpp clients can poll the netpp node at high frequencies without disturbing each other. The blue trace is a repeated poll of the full property tree, the red bars are the timed queries from a process viewer daemon. With no other disturbance, we get the occasional drops (e.g. at 45s, 101.5s) due to the queue running full


In-Field/System update

The default boot loader firmware supports self-programming over the cable. That means, the netpp_node can be supplied remotely with a new firmware image via a simple upgrade procedure over netpp. If the uploaded image is faulty, the system will fall back to the default boot loader. However, if the new design itself has errors, the system will be unable to recover  unless the reset button is pressed.

Test procedures

As the full model of this design is available for simulation, we can verify the system effectively against stress situations. In particular, network safety is of outmost importance. The test procedure check list of the dagobert SoC:


  • ARP and ping flooding
  • netpp packet performance test
  • Broken packet handling
  • Lost interrupt scenario (packet queue desynchronization)

Yet open

  • Jumbo packet flooding was tested, however support can not be enabled on this platform for the receiver queue. It is however possible under certain circumstances to generate (TX) Jumbo packets for experimental purposes. The performance gain is however minimal.

Extended RTOS support

Currently, the netpp node runs a simply bare metal main loop without particular RTOS functionality, i.e. all user code must be designed such that there are no blocking wait statements. Let me just put the FAQ together:

  • There is FreeRTOS and eCos support code for the ZPU architecture. However, I have no plans in going down that road, you’d be on your own.
  • A NuttX port is currently under evaluation and may likely be released in a few months time. NO PROMISES!
  • A simple ‘netpp OS’ with very basic task management is in experimental stage:
    • Guaranteed latency time from driver interrupt to queue handler task
    • ‘User space’ context switch when sleep() called
    • Very cheap context switches due to ZPU architecture improvements

Code size is an issue on this particular platform, larger programs (TCP stacks) need to move to the SPI flash overlay program space. Since this involves caching, the program timing is no longer fully deterministic and the RTOS functionality can only apply to program code running in the L1 memory.


Posted on

JPEG encoding on FPGA [revisited]

Although considered ‘ancient’ because invented in the early nineties, the JPEG standard is far from being dead or superseded. Its basic methods are still up to date for modern video compression.

For low latency image streaming, we have developed our own system on chip encoder solution ‘dorothea’ in 2013. It is based on the second generation L2 (a tag referring to ‘two lane’) JPEG engine, allowing JPEG compression of YCbCr 4:2:2 video at full pixelclock.

The ‘dorothea’ SoC is now superseded by the new ZPUng architecture, allowing more microcode tricks than on the previous MIPS based SoC. It is available as reference design ‘dombert’ (see SoC design overview) for UDP streaming up to 100 Mbps, optionally, 1G cores (third party) can be deployed as well for more throughput.

L2 example videos

These example videos are taken by direct capture (as coming from the camera) of the UDP video stream. The direct Bayer to YUV422 method is implemented in a microcode engine and may still show visible artefacts, also, color correction is not implemented for this demo. For the live videos, a MT9M024 sensor on the HDR60 development kit has been used. A bit file for the HDR60 kit is available on request.

Posted on

DMA Autobuffering techniques

For high speed DMA throughput where you’d want the least interruptions, several CPU cores on the higher end deploy scatter-gather style DMA engines to avoid slow memory copying. Let me give an example:

  1. Assume you would want to compose a network package and stream it out to a media access controller IP core (MAC)
  2. You have one Ethernet/IP/UDP/RTP header or alike and you are not planning to change it much during a streaming session
  3. Your payload might be in memory, but also coming from an external data port (such as a video data)
DMA descriptor scheme

The DMAA core (part of the cCAP families starting with ‘d’, like dombert) allows to set up descriptors in a special shared (and fast) dual port memory such that the DMAA engine can stream almost continuously to a peripheral. Likewise, descriptors can be set up for input streams (peripheral to memory), if required. The above image should talk. You start the DMA engine by writing the descriptor address into the DMAA_START_DESC register. Once the transaction has completed, an IRQ will be fired (which will for example allow you to update the sequence number of an RTP packet). Meanwhile, the DMAA engine will fetch the NEXT descriptor pointer (next()) and stream the payload data pointed to by ptr. You just have to make sure the IRQ routine does not waste too much time doing other things, if you use the same header in the following packet.

Note the control bits: an IRQ will for example only be fired, when the according IRQ bit is set. The EN bit tells the DMA engine to keep going after the current descriptor. The entire transaction stops when the last descriptor has the EN bit set to 0.

In the example for a MAC controller, you might want to append several chunks of data before actually issuing an Ethernet packet. How would the MAC know how long the packet actually is, without explicit (and timing critical) writing of length registers? This is taken care of by the FL (FLUSH) bit. Once a DMA transaction has completed and the FL bit is set, the packet is flushed in one go to the MAC and sent out immediately. The DMA engine waits until the data has been transmitted to the Packet FIFO and then resumes with the next descriptor.

In fact, this concept requires very little logic overhead and can run at high speeds even on small FPGAs (with a reasonable amount of memory buffers shared between DMAA and CPU core)

High speed streaming

When streaming high speed raw data, you might not want the CPU composing packets from external sources. Better stream them directly from the peripheral to the MAC the interleaved way. For the above Payload packet descriptor, the alternative data input method would be by setting the DP (DATAPORT) bit. In this case, the ptr and inc attribute of the descriptor are ignored and `count` number of bytes plus one are taken from the dataport input FIFO. If there is a premature end to the data, a CANCEL action can occur. The DMA will then stop (ignore the EN bit) and the user space program can react accordingly by checking if the DMAA engine is active. The effectively written number of bytes can be read from a DMA_CURCOUNT register and if necessary, padding action can occur.

Complex DMAA setup

Likewise, receiving packets of a priori unknown sizes is possible using this approach. A receive DMA IRQ handler just checks this register and fills in the effective packet size in the receiver packet queue which is later polled by ‘user space’ (this being the bare metal main loop, typically).

This way, data rates up to the theoretical maximum can be achieved. The rest is a matter of configuring the right packet and FIFO sizes. Below you can see a Wireshark packet graph for a regular packet burst (30 frames per second). As you can see, it’s cranked up to the maximum possible throughput during the burst.

Wireshark Packet burst


Advanced DMA capabilities are easy to implement in FPGA SoCs and make high speed transfers possible even with simple and slow CPUs.

Some reference applications:

  • Almost-Zero latency streaming of compressed Video over RTP (RealTimeProtocol)
  • Signal Analyzer and trace units (‘digital scopes’)
Posted on

netpp 0.5 release

Netpp 0.5 release notes

The netpp v0.5 release includes new functionality:

  • Proxy support: A netpp slave can connect to other slaves via netpp or other protocols
  • Support for the new ZPUng / MaSoCist setup (netpp on FPGA!)
  • More platforms added: esp8266/xtensa, avr, cc430, armv5, mips32
  • Improved dynamic property handling
  • Windows Python installer for Python v2.7
  • Java class wrappers
  • Added support for SWAP, modbus and MQTT (commercial)