Show HN: I made an Ethernet transceiver from logic gates

https://imihajlov.tk/blog/posts/eth-to-spi/

by ynoxinul • 4 years ago

460 85 4 years ago

imihajlov.tk

verisimi• 4 years ago

How is this fairly nerdy piece, that no one has commented on, top of hn? What am I missing?

tintedfireglass• 4 years ago

the HN algorithm is a bigger mystery than the universe itself

revskill• 4 years ago

No, i think most of its part is manual update database to set the rank of a post.

stavros• 4 years ago

It's super cool.

Myrmornis• 4 years ago

That it's impressive.

db48x3• 4 years ago

The very best things speak for themselves, and need no amplifying comments.

omeze• 4 years ago

Its been way too long since I touched a breadboard, but what software was the post using to simulate this? I would expect that the requirement for ~ns precision (microwave frequencies) would involve a lot more tedious calculations for designing the circuit.

ynoxinul9• 4 years ago

I used Falstad curcuit simulator (https://www.falstad.com/circuit/circuitjs.html) to try out some subcircuits (monostable, for example) and then tried them out on a breadboard.

Evidlo• 4 years ago

A mistake: you refer to C9 in the monostable section, but it should be C5

ynoxinul9• 4 years ago

Thanks, fixed it.

exmadscientist2• 4 years ago

Another mistake that doesn't deserve a top-level comment: the unit abbreviation for "second" is "s", not "S" (that one's the siemens = inverse ohm = former mho, which was a way better unit than the siemens, but that's another discussion...). Thus, "ms" and "ns" rather than "mS" and "nS". This may sound incredibly pedantic, and it is, but it's also one of the tests for attention to detail that I use when interviewing and evaluating an EE candidate, so, y'know, maybe worth learning to get right. Or not, life's short and building stuff out of discretes is more fun. Your call!

ynoxinul9• 4 years ago

Thank you, fixed.

unwind• 4 years ago

Hadn't heard of "mho" for conductivity, is that from when Bourne worked with electronics? :)

I would oppose a unit that starts with a prefix symbol (m=milli=0.001) on principle.

userbinator• 4 years ago

However, some hardware I have sometimes produced longer preambles

How long were they, and did they happen to coincide with short frames? It's been a long time since I've worked with 10M Ethernet PHY but I vaguely remember some non-standard implementations would lengthen the preamble instead of padding the "tail" of the frame.

ynoxinul9• 4 years ago

The preambles were longer by a few bits (even less than 8). I didn't check which frames excatly caused this, but it happened only with one old 10 MBit switch.

oakwhiz• 4 years ago

It's possible that some PLLs might fail to lock in within a normal amount of time to acquire the full preamble. In which case varying the length of the preamble slightly shouldn't negatively affect receivers that are well designed to tolerate unusual conditions.

mgsouth• 4 years ago

Cool! Have you submitted to Hackaday?

55555• 4 years ago

Wow the tech talent in Tokelau is insane!

charcircuit• 4 years ago

What is the benefit of using all of these different components over using an FPGA?

ynoxinul9• 4 years ago

It's a fun project. I feel like building stuff with logic gates.

meltedcapacitor• 4 years ago

It is hardware poetry.

(What is the benefit of rhymes, meter, etc over using prose?)

PostOnce3• 4 years ago

It's worth your while to buy another domain. You could keep the .tk running.

Some sites shadowban .tk, Reddit probably still does. Search engines probably downrank it. That's because it's free and has been used for a lot of spammy stuff.

This makes it harder for people to share and harder for people to find. That might just put a damper on your hobby, but it might also mean you won't get random emails about jobs you might want. (If you're interested in that.)

I only mention it because I learned that the hard way many years ago.

moffkalast5• 4 years ago

Is it worth it though? Renewing a domain is yet another bottomless pit you throw money into, along with server hosting. For a site without ads or other revenue it's just more expense.

actionfromafar• 4 years ago

Static hosting can be found free on sites like gatsby.

samjmck2• 4 years ago

imihajlov.com would cost around 10 USD per month and as another reply mentioned, static site hosting is free. I guess it's a bottomless pit but it's "only" 10 USD and it's nice to have.

TaylorAlexander3• 4 years ago

Yeah but what if it’s a cool domain? Then it’s a cool bottomless pit.

PostOnce3• 4 years ago

It very well might be, if you value publicity for your hobby, the greater reach might be worth it to you. If you can get 10,000 readers instead of 90 readers because your links are shareable...

If we think about potential dollars, though:

The bottomless pit isn't very deep, it's a few dollars a year.

If it nets you one single short-contract offer (nevermind a job) or worthwhile contact because someone found your work, it'll pay itself off many, many times over. Let's say one person found him and offered him a $3000 contract to do some electronic thing for a week, well, then the cost of the domain and hosting won't matter for the rest of his life.

The mere fact that this guy has an oscilloscope out means he has skills that someone might want to contact him about in relation to his blog.

Limiting yourself for want of a $10/year domain name is probably not a great idea, and again, it's something I learned the hard way.

1kurac• 4 years ago

The thing with TK is, it costs nothing upfront, but the internet is bursting with countless reports that if your site eventually becomes popular, TK domain admins might outright snatch it from you for their own ads, without the possibility of seeking retribution as it's all in their TOS ... https://getfreedomain.name/domain/tk/

teddyh• 4 years ago

I know of exactly one site which had a .tk TLD. Here’s why they switched to a .com in 2005:

https://tailsteak.com/archive.php?num=388

“tailsteak.com

Alright, so I'm finally fed up with Tokelau.

Don't get me wrong, it's a nice island, but they just aren't answering my mail.

I suppose most of you have noticed, by now, the fifteen-second ads that present you with beautiful women and fish when you surf to tailsteak.tk. Those are not my ads. I do not obtain revenue from them. Tokelau's domain name referral service just started putting them up there without so much as a by-your-leave. I have contacted their tech support and enquired if, perhaps, they might consider removing them for customers willing to pay a certain amount. They have not responded.

Of course, I have had access to tailsteak.com for some time now. So henceforth, I will be directing my viewers there. It's the same site, the same host, and, in truth, the .tk address has been sending you there for months. But now it's official. Note the change in title graphic:

tailsteak.com”

marginalia_nu• 4 years ago

Yeah. Cheap TLDs are very strongly correlated with spam.

Although I think .icu and .xyz are far bigger offenders than .tk these days.

psychphysic• 4 years ago

I love this stuff, well over my head but amazing read and so grateful for the scope captures!

testermelon• 4 years ago

This is sick. I also read their post about the making a discrete computer and didn’t expect them to be also writing a compiler from scratch. Something I’ve only been dreaming about.

mikewarot• 4 years ago

So all I need are link pulses to fake the other end of an Ethernet connection to keep it alive?

This makes building a data diode much simpler than I thought it needed to be. Thanks!

ynoxinul9• 4 years ago

Yes, but this is only the case with 10BASE-T. Modern standards use more complicated Fast Link Pulses.

kevin_thibedeau• 4 years ago

Also applies to 100Base-T which is still a usable solution for basic networking.

birdyrooster• 4 years ago

I love these people bc they do all the esoteric stuff so that I don’t have to.

greenyoda2• 4 years ago

This is a wonderful project! I'm old enough to remember the days when Ethernet adapters for commercial computers were made of individual logic gates. The Ethernet adapter in the VAX 11/750 I used in the early 80's was two large Unibus boards full of TTL logic: http://gunkies.org/wiki/Digital_Ethernet_UNIBUS_Network_Adap...

aswanson2• 4 years ago

That is crazy, the amount of space a network adapter took up.

db48x3• 4 years ago

And there were a few more boards for the disk controller, and another couple for the display controller…

rasz3• 4 years ago

First PC network card was only slightly smaller :)

3Com 10Mbps Ethernet 3C500 aka IBM Ethernet (IE) aka EtherLink from October 1982

https://static.wixstatic.com/media/a03cac_005e3e9eb62b47c292...

Gordonjcp3• 4 years ago

The one that superseded that, the DELUA adaptor, was powered by a 68000 CPU that was generally faster than the machine hosting it :-D

tyingq• 4 years ago

This blog post dives into another mostly TTL logic Ethernet board from the Xerox Alto: https://www.righto.com/2017/11/fixing-ethernet-board-from-vi...

It has a nice high level diagram of what each group of TTL chips is doing too:

https://lh3.googleusercontent.com/-DpUyda3g2Yk/WgB7G_7o54I/A...

amelius• 4 years ago

From Wikipedia:

Manchester code was used in early Ethernet physical layer standards

So, not in later standards?

detaro• 4 years ago

Later standards use more efficient codes that map longer symbols to DC-free codes, and thus are more bandwidth-efficient. (8b/10b, 64b/66b codes, where manchester is basically a 1b/2b code)

Aromasin• 4 years ago

Interesting timing for this post - I'm studying this for a new job at the moment and building myself my own little Wikipedia in order to learn it. 10BASE uses Manchester Encoding, 100BASE uses Non-Return-to-Zero, Invert-on-one (NRZI) encoding and Multiple Level Transition (MLT-3) with 4B5B translation, 1000BASE uses 8b/10b Encoding (8 bits for data, 2 for ECC).

More recent speeds all tend to use PAM4 encoding, meaning there's 4 voltage levels instead of 2! There's a great article on it here: https://blog.samtec.com/post/understanding-nrz-and-pam4-sign...

alexnewman• 4 years ago

classic hackernews. We need more of these

jtlienwis• 4 years ago

Look on Hackaday for "Why do you need to use decoupling caps". I am not seeing any talk of decoupling caps in any of this persons writeups. Maybe they are there and I am just not seeing them. It used to be when designing with LS ttl you needed one per chip to make sure switching noise did not get onto your power supply rails. Otherwise, a great project for learning.

ynoxinul9• 4 years ago

Do you think they are worth mentioning? My post isn't a tutorial.

bogantech• 4 years ago

The decoupling caps are there in the schematics shown alongside the symbols for the IC power inputs which is a common way of doing things.

IDK why something as fundamental as decoupling caps should be mentioned here at all though, anyone sufficiently knowledgeable to make use of this information would know that they're required already.

0xmarcin• 4 years ago

What a mervell of retro-computing. I think the author should definitely contact Ben Eater, as this is an excellent material for the next video.

Aside from this, Arduino shields for Ethernet are available for quite some time, but they often contain the more advanced chip than the atmega328p itself. Here OTOH the author builds from the ground up. Simply amazing!

That being said, I think most of the ppl that are serious about their arduino networking will use CAN as their network of choice (https://en.wikipedia.org/wiki/CAN_bus).

bjt2n3904• 4 years ago

Oh wow.

I had thought this was, "I simulated it in a high level HDL and synthesized it to gates".

No, this is... I built it with physical chips. Brilliant build and write-up!

Edit: This is beautiful for a second reason... The lack of regard for doing things "correctly". The ugly clock circuit is horrifically unreliable... But just reliable enough for it to still work.

It reminds me of Muntzing.

https://en.wikipedia.org/wiki/Muntzing

systems_glitch• 4 years ago

Nice work! Now do thicknet :P

bogomipz5• 4 years ago

Wow, this is fantastic! I had a couple of questions. The author states:

"According to the 10BASE-T spec, each frame starts with a fixed synchronization sequence of 64 bits, first 62 of which are alternating ones and zeroes and the last two are ones."

I believe the intention of the "preamble" is to provide clocking and synchronization for the circuit so that it knows where to find the start frame delimiter and subsequent fields. Is that correct?

My other question is at what point does the circuit drop frames that are not intended for it? Does that happen in layer 1 or layer 2? I realize this happens less in modern switched Ethernet networks but it is still possible for a host to get an Ethernet frame for where the destination MAC address does not match any of those burned-in addresses on the Ethernet adapter. Does the transceiver always read the entire Ethernet frame even when the destination MAC address field it reads is incorrect? The maximum Ethernet frame size is 1518 bytes which seems like a lot of data read if the MAC address is not valid for the host.

ynoxinul9• 4 years ago

The preamble is intended for synchronization, the last two consequtive ones indicate the start of the data.

My circuit doesn't drop any frames, it simply converts serial Ethernet signal to SPI. FCS and MAC checking is done in software after receiving the whole frame.

You absolutely can get frames not addressed to you. Wireshark lets you do that. I debugged my adapter this way: I connected it to a dumb switch together with my computer and ran Wireshark on the computer which captured all frames regardless of their intended destination. On Linux you can even disable FCS checking and receive all rubbish from the wire.

bogomipz5• 4 years ago

Thanks, I had another question after a second reading of your post:

"The first bit of an Ethernet frame is always one, which means there might or might not be an extra edge in the beginning depending on the idle state of the amplifier. This extra edge, if it comes, needs to be filtered out."

Practically speaking what is the cause of the "extra" edge here? Would it be that the last bit of the FCS from the previous frame caused it to be a 1 and then the transceiver is immediately getting another edge from a preamble on a new Ethernet frame? Or am I confusing bits and edges?

zamadatix3• 4 years ago

Imagine you had 2 transceivers you wanted to bridge - if the transceiver dropped all unicast MACs but its own indiscriminately then it would not function properly. There is also a litany of other features that would not want this kind of offload always running (general MAC spoofing, packet capturing for debug/troubleshooting purposes, VMs, some IP redundancy protocols that keep the virtual IP's MAC consistent after failover, and the like).

Offloads such as this or other more advanced ones are implemented as optional hardware features (if at all) controlled by upstream signaling. This is usually a driver signaling things over a PCIe bus on modern systems but it can be anything including just "upstream logic gates". That way if it's a normal client-only PC not running e.g. Wireshark or VMs the CPU doesn't have to be interrupted for the types of traffic you mention but the other use cases can still work too.

bogomipz5• 4 years ago

Thanks, that makes perfect sense about things like VRRP and proxy arp, indeed they wouldn't work if the transceiver filtered on the destination MAC field of the Ethernet frame. I'm guessing this is exactly why this stuff is handled at L2 and in the kernel module for the NIC? I didn't understand your first example though:

"Imagine you had 2 transceivers you wanted to bridge - if the transceiver dropped all unicast MACs but its own indiscriminately then it would not function properly"

I'm not understanding why bridging makes a difference. In the case of a bridge a host would send out a broadcast i.e a frame with destination MAC ff:ff:ff:ff:ff:ff which is valid for all Ethernet adapters, this is similar whether it's a switched or bridge network no? Maybe I'm missing something obvious?

"Offloads such as this or other more advanced ones are implemented as optional hardware features (if at all) controlled by upstream signaling."

Which offloads are you referring to here? Just general TCP offload engine stuff? How does that relate?

rasz3• 4 years ago

Sync pulses are for PLL to sync with transmitter clock. Floppy sectors also start with a train of 64 sync pulses https://info-coach.fr/atari/hardware/FD-Hard.php#FDC_PLL

Show HN: I made an Ethernet transceiver from logic gates

Related Stories