I am not sure where I wanna use this W5500 circuit, but the nice thing is that KiKad makes this a module I can re-use. This actually is the reference schematics.
Fiber Modules
Fiber modules have been around for some years. The picture below show a 14.- USD 8 x Ethernet/2 x Fiber Mux/Switch that can be found on AliExpress etc. I have also seen smaller modules and 16 x Ethernet modules in the same price range – also boxed versions.
The interesting thing about these is that the Fibre links in theory can stretch a cable up to 100km. As I mentioned this has been around for decades, but it impresses me to see these modules down below 20.- USD in cost.
My own PLC uses high speed SPI as backbone and I already have Ethernet through Raspberry PI or ESP32 Board, but I fancied drawing a dual ethernet solution based on Wiznet W5500 as well. I have used this chip before. Ethernet and Fiber will extend the backbone to cover kilometers of wired net then needed. I am also a bit interested in using Ethernet as backbone all over, but space is a bit limited on the current Hat format. If I am to upgrade to Ethernet as backbone I will need to increase size, so having an Ethernet adapter is a compromize.
The ESP32 Ethernet module contains Ethernet and Wifi, but the SPI on it reduces backbone speed to ca 9Mhz as Slave. I can live with that, but I would prefer a higher speed. It is however challenges involved in using SPI the way I do, and a pure Ethernet backbone have a lot of advantages. But, is has one disadvantage – added space need on my boards + you need the switch in the backplane. And don’t forget that we also have CAN that can stretch over 1km.
Another option is to upgrade XPortHub and replace the SPI/I2C ports with an Ethernet port. I could just use magnets and a different, smalle connector etc. I have a bit of space problem here, but it can be done if I either sacrifice TF card or move TF card and/or battery to back side of card.
C#/.Net Dark Theme
Working on HMI in C#/.NET is fun, but it is a large task to get dark themes right. Below is a list of components that I want to complete:
| clEditCtrl | Simple text edit control. |
| clHScrollBar | Horizontal Scroll Bar. |
| clObjectEditor | Advanced grid alike editor to edit an “object” with a tree of items. |
| clPopUpMenu | Classic pop-up menu. |
| clShapeLabel | Label with titled corner options. |
| clSmallButton | Small, flat button with image symbols. |
| clTabBottonCtrl | Advanced button with sub-buttons. |
| clTabStrip | A list of buttons located on an edge. |
| clVScrollBar | Vertical Scroll Bar. |
| clMenu | Classic Horizontal Menu |
| clCaption | Dialog title caption |
| clToolBar | Horizontal bar for various sub tools |
Other components will be added as needed, but I want to complete the list above so I can move on with other things for now. I decided to not do a drop in replacement of all components, many of the components above also add advanced functionality to be able to quickly create an advanced HMI.
I have tried various schemes, but I ended up sub-classing from UserControl and re-create things from scratch on most components. Luckily I could use the edit control with a few tricks, because that alone is a lot of work. To make this work I had to add color properties as follows on all components.
| BackColor | Most components have this |
| ForeColor | Most have this as well. |
| BorderStyle | Need to override BorderStyle |
| BorderColor | Very few components could control border color from the application. |
| BackColorHover | Background color if mouse hover over the component. |
| BackColorShape | Used only on shape controls where I need to “normal” background colors. |
| BackColorSelected | Used on controls like push-buttons to signal state. |
I started doing this on Qt as well earlier, but I concluded that C# was so much easier to work with and as performance issues I had earlier are gone I moved back to C#/.NET. I might go back and re-visit Qt at some point.
CANopen/SIIS2 – Part 4, PDO Map
A PDO differ from a SDO in the sence that the COB ID is the message ID. In a single system you have 8 x 127 messages that can be PDO and only needs to be know by consumer and producer. A simple PDO Map can map directly to memory variables, but a more proper implementation will do so indirectly through SDO Index Map.
In a Static PDO Map this is assigned by Firmware at start-up, while in a dynamic PDO Map this can be configured with separate messages.
In terminology you have PDO, MPDO, RPDO, TPDO, but to keep it simple only use PDO. RPDO and TPDO are separate groups as you have 4 x RPDO and 4 x TPDO groups – how they are used is up to the device profile. MPDO (Multiplexed PDO) is something I have never used, so my knowledge about that is a bit thin (yet).
CANopen/SIIS2 – Part 3, SDO Index Map
SDO is the simplest part of CANopen that can be compared to Modbus. A SDO message is addressed to a node and consist of Index, Sub-index and 1-4 byte variable. A common implementation is to have indexes in an array with pointers to the actual variable and use bsearch to find index/sub-index. How you implement this is not part of the standard, so you could just implement a very large switch etc, but a table is recommended.
In fact, most stacks have a Design Tool that will allow you to generate this table in source code and the associated EDS file.
CANopen/SIIS2 – part 2, stack UML diagram
I am considering using CANopenNode to create a full SIIS2 stack, but before I jump into someone elses work I like to think through how and why I want CANopen build. Below is a simple UML diagram.
This diagram lack Bootloader and Object Dictionary, but I simply created classes for each bulk of work that needs to be done.
CANbus. I need a driver outside the stack. This will integrate to various HW as well as adapters for Windows etc.
CANopen. We definitely need a CANopen with all main interface functions. A simple class to be used to create a stack and then add CANModules to this as needed. We can createa a CANopenSlave and CANopenMaster, but also provide a custom implementation.
The rest of the CANopen stack is easily divided into groups of work using CANModule as base. CANModule can carry common functions while each of the SDO, PDO, LSS, NMT etc carry their respective messages with Processing, Encoding and Decoding. Some of the blocks like SDO, PDO, LSS and NMT should be sub-classed because Client/Server, Producer/Consumer, Master/Slave have different implementations of the same messages. So can EMCY, SYNC and Time, but as these are smaller I am not sure I bother.
Creating a C++ layer like this also allows for the user to sub-class and easily modify the stack without changing the base code or dealing with nasty #ifdef code.
Heartbeat is actually part of NMT with separate schemes for Master and Slave.
This is how I would like to divide code up in C++ and if you look at CANopenNode you notice that these building blocks more or less exist in C code – this was accidentally.
CANopen Bootloader will need to be created from scratch.
Object Dictionary will need to be re-done. The onle used in CANopenNode is not sufficient for what I want to do.
That said – I am not convinced I will jum onto this train – just toying with the idea for now. Wherever I write my own from scratch or use CANopenNode is of no consequence as I can easily write a C++ wrapper on top of CANopenNode.
CANopen/SIIS2 – Part 1
The industry around me is a heavy user of CANopen (SIIS2) which is the “Subsea” version of a CAN protocol. CANopen is similar to Modbus, but it offer more functionality. It is also a more complicated protocol stack that tend to cost from 5-15K USD depending on content.
I have written CANopen protocols in the past so I am toying with the idea to write one in C++ for embedded systems, while I am also happy to see that CANopenNode have changed license from GPL to Apache 2.0. CANopenNode looks like a nice implementation that cover ca 50% of a full stack.
The modbus standard is available while CANopen standard is close to 300 cryptic docs that you have to pay for. The later makes it difficult for people to write open source or promote the standard.
The main functionality is very similar between Modbus and CANopen – in modbus you read/write registers, while in CANopen you read/write objects which for most parts are 32 bit variables.
Both CAN and RS485 (that often is used for Modbus) uses a twisted, differential wire. The main difference is that CANbus is more difficult since it uses arbitration.
On Modbus you have a request/response protocol meaning Master send and wait for a Slave to respond.
On CAN all devices can send whenever they want using Arbitration. The 11 first bits in the header is arbitration field and basically the message with the lowest number “win”. This requires good timing and an aligned bus to work. The drawback is that the payload on classic CAN is 8 bytes. More modern bus systems like CAN FD and FlexRay that also uses arbitration have bigger payloads, but few MCU’s support these yet.
Modbus is a few messages and job done – it is a very simple protocol to wrap up.
Modbus have high latency, while CANbus support burst – a technique to just send messages as fast as you can. The Burst technique more than compensate for the header and low payload – CANopen is far more efficient in data transfer than your would expect.
Modbus based on RS485 can support 2Mbps or 10Mbps serial lines, CANbus is max 1Mbps.
CANbus have a High Speed and Low Speed/Fault Tolerant version. The FT version means the transceiver can drop to half-duplex and continue to work with only one of the twisted pair wires as long as it also have Ground. The FT version is to my knowledge only used in SIIS2.
CANopen is a large job consisting of several components:
Bootlolader. This do not exist in Modbus, but CANopen have it’s own bootloader with a defined boot sequence where a device ask a Network Master (NMT) if it should start or download new firmware.
SDO is the main protocol to send generic data up/down. It support single variables and blocks of data both ways.
PDO allows you to send proprietary messages from node to node using all 8 bytes for data. SDO can only use 4 bytes, while PDO messages can be packed.
NMT (Network Master) perform start up sequence and monitor the network.
LSS (Layered Service Settings) consist of some 33 message sequences that control the node and its services.
And you also have the smaller components like Heartbeat, EMCY, SYNC and TIME.
The main part of a CANopen is however the object dictionary – which is similar to modbus registers. This is the list of objects and a huge portion of the CANopen standard is dedicated to describe how various devices should look like – what interface they should have.
CiA402 define a motor controller, while CiA443 define several standard Subsea components as examples.
The standards for CANopen have an extremely low quality – errors and inconsistent, little description to help understand usage etc – I have worked close to two decades with telecom protocols – actually writing protocols like ISDN and SS7 – and CANopen is the worst standard I have ever seen – it is a mess of a standard.
CANopen is NOT difficult as such, but understanding the so called standard in an uniform way and making sure devices integrate properly have turned out to be a lot of work.
Terminology like Master and Slave stack is used. CANopen have a few Master components, but it is actually a distributed network where some of the nodes takes on resposibilities. The work “Master” do however also means you have a device that is capable to track other devices – it need tables over the other devices etc, while a “Slave” or a “Device Stack” only need to know about itself.
For those new to CANopen I will recommend starting out with CANopenNode – this is open source C code under Apache 2.0 license. The stack is still work in progress, but I had no problem following some message sequences and noticing what was working correct/incorrect – which at the end means the code is mantainable.
Dark Theme in C#
Many modern applications use custom theme/styles, but this is actually far more difficult to implement than you would expect on Windows. Windows have a lot of standard GUI components where you can control most, but not all colors and styles. So, if you want dark theme you either need to buy a library or start making custom components that actually control all colors and styles. I decided to do the later. And while at it I also add new super components to make life easier. My objective is a dark theme not so unlike the one used in Visual Studio itself. I decided to code desktop apps in C# simply because Forms have improved their graphics and C# Forms is easy to work with.
This simple app is my component scratchpad showing some of the components I have made so far. You see small buttons, scroll bars, labels and edit fields. On bottom right you see a classic Windows combo box and as you can see I can’t control border colors yet. Looking at the edit controls (top right) I almost got it right using a standard component, but I could not control the border color. What I did was to use a edit component without a border inside a custom control where I override BorderStyle. I am realy pleased with this one because an edit text component is a bit of work.
The next control I had to work on was the tab strip. I have to modify the close cross a bit, but again it is a custom control with a small button and custom label component. As you can see I still have to do the menu and windows title fields on top – in fact I want to move the menu into the caption bar just like in Visual Studio to save space.
This is the beginning of my own Object Editor. I tried hard to avoid having to make this, but at the end I decided to jump into it. First of all this is a complex component as it is a simple grid with in-line editors and at the end it simply was to much that I could not control. But, having made the basic components properly it is easy to re-use them in a grid like this. I have not coded attributes properly so it just use reflection to grab a few edit fields for show yet.
This is a lot of work, but it is also fun. I want a proper HMI on my systems and I have made so much electronics that it is due time I start on the top-side part. C# also support Linux, Android and OSX. I actually ported a Forms application to Ubuntu once with just some minor changes. It will be a bit more work to port this HMI framework, but one step at the time. At this time I want to get the designer working and C# is the fastest way forward.
BasicPI Designers
UML is a well accepted standard for modeling complex software solutions, but it has never reached executable accuracy. The standard lack a technique that allow executable logic, which is exactly what PLD does. This is key to having a tool with 1:1 between diagrams and code.
I started modeling with Entity-Relationship Diagrams years ago, and these diagrams still have great value on physical database designs. ER diagrams are also a sub-set of class diagrams on the front-end, so they are easy to include.
HMI design is crucial if you deal with HMI. Starting this designer task I wanted dark-theme, and that means I have to take control of every HMI component to control colors anyway. Creating an HMI designer is on the front-end only about drawing symbols on the screen – far easier that the complex diagrams with rules.
Putting an ER designer and HMI designer together is alone a powerfully tool as you create database applications in seconds. This is what classic CASE tools used to focus on. Having this capability is nice, but only a side-effect of what I want.
PLD is a new technique combining two very popular techniques. Classic flow charts to express executable logic has been here since the beginning. SDL (Specification and Description Language) is still heavy used in telecom. PLD build on these two for graphical language and extend with techniques to have an executable specification. A diagram that in a few second describe what the logic is.
UML State Diagrams sneaked in because I liked the concept and it fills a gap between PLD and System Diagrams. PLD can be used for system diagrams, but state diagrams are more efficient.
As we also work in object oriented languages a class diagram is a must. This looks like an extension of a ER Diagram, but while a ER Diagram represent physical tables in a database (the actual container) a class diagram represent a definition of an object that needs to be declared – it is an advanced datatype. I use class diagrams heavy as I code in C++ and C#.
At the end we want to press a button and generate code. The concept is that you draw diagrams, press a button and your up running. But, it is good reasons why attempts on achieving this has failed in the past – lets see if it actually work this time!
Coding with UML State Diagrams
Some of you will remember that I planned a “System Diagram” in Plain! That was a block of configuration code that tied event on modules together. This is exactly what I plan to use UML State Diagrams for.
In this diagram we generate a Temp Changed event on one device that is sent to a Display on another device. UML State Diagrams are excellent for visualizing this event flow, but also capable to showing more detailed flow. What UML State Diagrams can not do is to visualize logic very well, so for that I use PLD or simply direct source code.
Lets forget about the internal logic in the two devices for now. We have one device that sample temperature 100 times per sec, a second unit that is capable of receiving and displaying temperature. The diagram simply say that as the temperature event is generated it will be sent to the display.
Temperature
- – time stamp
- – temperature
In this example we use easyIPC, so all the code we actually need is a “Wire” command at startup that is sent from “System Master” to Temp. Device:
- Look up address of Display Device
- Send Wire Temp_Device.Temperature to Display.TempInput
This will cause the Temp_Device to start sending Temperature messages 100 times a sec to Display. Now, sending temperature 100 timer per sec is a bit overkill, so we modify the diagram a little. In this case I add a smoothing function in. Smoothing is a pre-defined function in code, so all I need to do is to create the state and route temperature in and set smoothing Window to 2 sec with 1 sec output frequency. Smoothing is in effect average, so what this does is that we send the average temperature for the last 2 seconds every second.
Generated code will be something like this on the System Master at start-up:
- Look up address of Display Device
- Look up address of Smoothing function
- Send Wire Temp_Device.Temperature to Temp_Device.Smoothing.Input
- Send Wire Temp_Device.Smoothing.Output to Display.Tempinput
This example is highly simplified, but it illustrates top level UML State Diagram usage that can’t be covered that easily in other techniques.