Ethernet bit rate

[radiolistener]

redkitedesign

it uses 2 voltage levels for transmission, so it can be handled with digital GPIO.
But 1000BASE-T has 5 voltage levels.

[m k]

I happen to have The New Webster Dictionary '67.

Since the definition is not very old it seems that you can demand some leash.
But only if you're out of the loop.

[redkitedesign]

redkitedesign
it uses 2 voltage levels for transmission, so it can be handled with digital GPIO.

Nope, E1 (and T1 too by the way) use 3 voltage levels. And you show me a GPIO that produces negative (wrt GND) pulses.

[coppice]

redkitedesign
it uses 2 voltage levels for transmission, so it can be handled with digital GPIO.

Nope, E1 (and T1 too by the way) use 3 voltage levels. And you show me a GPIO that produces negative (wrt GND) pulses.

There is a reason for the "3" in HDB3.

I think there was a variant of E1 which did actually use a 2 level code. I don't think I ever saw it used. HDB3 is the normal code for E1. Three level codes are a pretty good way to deal with the DC balance problem.

[radiolistener]

Nope, E1 (and T1 too by the way) use 3 voltage levels. And you show me a GPIO that produces negative (wrt GND) pulses.

In that case you're just use several GPIO as DAC (mixed signal circuit).
The same you can use 1-bit GPIO with PWM and analog low pass filter to generate analog signal, but this is not digital, but mixed signal DAC circuit. Isn't it? 

Digital element can be used in mixed signal circuit, but it doesn't means that such circuit is clean digital, because it works with analog signals. The example of such mixed signal circuit is a classic parallel DAC. It uses digital outputs to generate analog siginal. But it doesn't means that analog signal from the DAC is digital.

[Berni]

All digital is actually analog once you look at it closely enough.

For example high speed interfaces like PCIe do just send 1 or 0 over the wire. However due to the kind of speeds it runs at the signal path between two chips can no longer be guaranteed to be nicely flat in frequency up to these sort of GHz bandwidths. This can make the eye diagram close up and introduce bit errors with certain aggressive bit sequences. So one fix that a lot of these interfaces use is pre-emphesis, that way the transmitting side increases the drive voltage near the edge transitions, so that extra high frequency content is put into the signal, once the signal gets to the receiving end the PCB traces have distorted the signal back to being a nice clean square wave again.

Then there is also the aspect of timing. With high speeds you can have skew due to having slightly different time of flight on each wire, so you need to measure the skew and shift your sampling point over to avoid sampling its value inside a signal transition, so it is not purely digital if there is a analog aspect to the timing of the signal. Much like PWM is not really digital, it is just a square wave signal that represents some analog average value (but the difference is the value it represents it analog)

Flash memory also uses multiple levels to store more information inside one flash cell, tho we don't talk about it as being "analog flash memory".

In my opinion these all fall under being digital because they represent information on both the TX and RX side as discrete levels with a fixed resolution. Technically this means that connecting a 24bit DAC over to a 24bit ADC can be described as a digital link utilizing PAM16777216 modulation. However your bit error rate is going to be horrendously bad, so you better have some very very robust ECC on the data.

You can argue for being digital or analog depending on what you want the answer to be.

[radiolistener]

In my opinion these all fall under being digital because they represent information on both the TX and RX side as discrete levels

gigabit PHY uses the same pair for RX and TX simultaneously, it can transmit and receive simultaneously in both directions through the same wire and their levels are not fixed, the levels can vary depends on a line length and transmitter gain which is setup by PHY chip during link negotiation when you attach the cable to the port. It carry discrete signal, but it is carried through analog medium over variable analog signal.

For gigabit pair there is needs for at least analog signal substraction and analog signal addition in order to TX or RX something. This is can't be done with digital circuit.

In short, PHY just establish connection through analog line with analog signal and transfers digital data over that link.

You can say that analog RF spectrum is digital because some device can transmit some digital data through it, but it is not digital. For example, the digital line can be used to transfer constant signals which changes very rarely, but you can't do it with gigabit pair between two PHY, because this is analog line and it requires analog signal processing for TX and RX something...


Did you seen RF transformers on the path of digital signal in digital circuit? It will break digital communication. But gigabit pair is connected through transformers (which are placed in RJ connector).

I just categorize digital line as a line which has a fixed valid voltage ranges for a fixed digital levels 0 and 1, other voltage levels are not allowed for digital line and should not appears for a period which is longer than some time limit which depends on digital line bandwidth. If signal doesn't fit into this scheme then it is not digital, but analog and needs to be treated with analog circuits.

[wasedadoc]

I just categorize digital line as a line which has a fixed valid voltage ranges for a fixed digital levels 0 and 1, other voltage levels are not allowed for digital line and should not appears for a period which is longer than some time limit which depends on digital line bandwidth. If signal doesn't fit into this scheme then it is not digital, but analog and needs to be treated with analog circuits.

You can categorise as you want but I'm reminded of the proud mother at her son's graduation ceremony from military college.  As they all marched past during the passing out parade she turned to another lady beside her and said: "Look at that. They are all marching out of step except my son".

[ve7xen]

I just categorize digital line as a line which has a fixed valid voltage ranges for a fixed digital levels 0 and 1, other voltage levels are not allowed for digital line and should not appears for a period which is longer than some time limit which depends on digital line bandwidth. If signal doesn't fit into this scheme then it is not digital, but analog and needs to be treated with analog circuits.

This is called a Non-return-to-zero level (NRZL) code.

Not only is it not the only digital coding scheme, it's not even the only binary digital coding scheme. For example, NRZI doesn't fit this definition, but is binary.

Digital lines convey digital data.
Analog lines convey analog data.

Both can have line codes and use modulation that changes the representation of the signals on the wire from their 'base' representation. Whether it's a simple binary NRZL or a complicated scheme like 10GBASE-T, the signal is continuous because it exists in the continuous (for our purposes) real world, and therefore needs a definition that works in the continuous domain. Either this makes all signals are analog, regardless of what they encode, or it means that what is encoded matters more than the continuous nature of the signal.

[radiolistener]

Digital lines convey digital data.
Analog lines convey analog data.

Then, I have two questions:

1) Can you categorize output from DAC? This line (named as "analog") convey digital data which is sent on input of DAC in binary form and discrete in time.

2) Can you categorize output of ADC? These lines represented in binary form and discrete in time convey analog data which is sent on analog input of ADC.

Regarding to 1000BASE-T, it's signal is produced with DAC, analog amplifier with programmable gain and passed through transformer like usual analog line. This line cannot be processed with digital input. It needs analog signal pre-processing and then needs analog to digital transformation before it can be processed in digital domain. It transfers continuous analog signal which convey both - RX and TX digital data simultaneously. And of course it is processed with analog circuits such as amplifiers. And this is why I can't accept "digital" type for that signal line.

[Siwastaja]

Digital lines convey digital data.
Analog lines convey analog data.

While some will keep disagreeing with this, that is the only useful definition, which is why I like it. The other technically correct possibility is to define: "every physical signal is analog, the concept of digital signals does not exist at all". While possible and correct, it's stupid IMHO. What keeps amusing me is radiolistener's inability to choose either of the technically robust definitions, and instead end up with an internally incoherent opinion which he somehow fails to see even though it's so blatantly obvious to everybody else.

[radiolistener]

every physical signal is analog, the concept of digital signals does not exist at all

there is no concept that digital signals doesn't exists at all.

The digital signal is a signal discrete in time and represented as a stream of digits, such digital signal is synchronized with master clock and needs to have valid state at the clock event. Valid state means that it should have stable and valid voltage level defined for digit in a specific digital signal standard.

On the contrary, analog signal is continuous in time and don't have valid ranges for some digit, it can have any voltage level in a working range. It also doesn't have a clock, because it is continuous.

Technically you can name one way PAM-5 stream as "digital", because it is represented as a stream of digits and discrete in time because it is synchronized with clock.

But the signal on a pair between two 1000BASE-T PHY is not just simple PAM-5 with a clock.

There are some difference:

1) there is needs digital to analog and analog to digital transformation, because PAM-5 encoding requires more than 2 voltage levels but digital domain operations can be done in a binary representation.

2) there is need to recover a clock from continuous signal representation on analog line, because there is no clock lines between two PHY

3) there is need for analog signal processing such as amplification/attenuation, analog add/subtractions, common mode currents rejection, echo cancellation, etc.

4) there are two streams on the same 1000BASE-T pair simultaneously, they flow in opposite direction and they are not synchronized with each other, so there is analog sum of two PAM-5 streams and they are clocked from different asynchronous clock sources.

As you can see, it has too many analog signal operations and the sum of all these analog signal operations cannot give you a chance to treat this signal as a simple stream of digits discrete in time. So, you cannot name it as digital.


When you saying that this line is "digital" just because it transfers digital data, this is a big simplification which omits many details which is actually very important for processing such a signal.

Another example, did you hear about HART protocol? This is a hybrid analog+digital protocol which allows to send digital data over analog 4-20 mA line which already has analog signal from analog source. Can you say that 4-20 mA analog line is digital just because it allows to transfer digital data?

[alm]

The digital signal is a signal discrete in time and represented as a stream of digits, such digital signal is synchronized with master clock and needs to have valid state at the clock event. Valid state means that it should have stable and valid voltage level defined for digit in a specific digital signal standard.

[...]
1) there is needs digital to analog and analog to digital transformation, because PAM-5 encoding requires more than 2 voltage levels but digital domain operations can be done in a binary representation.

2) there is need to recover a clock from continuous signal representation on analog line, because there is no clock lines between two PHY

The recurring logic flaw in your reasoning is the following reasoning:
Given circuit A and circuit B, if circuit A is a digital circuit and has property X, and circuit B does not have property X, then circuit B is not digital. This is a fallacy.

There are synchronous digital signals (e.g. SPI). There are also asynchronous digital signals (e.g. RS-232). There are binary digital signals, and there are also digital signals with more levels. One does not exclude the other.

[radiolistener]

What keeps amusing me is radiolistener's inability to choose either of the technically robust definitions, and instead end up with an internally incoherent opinion which he somehow fails to see even though it's so blatantly obvious to everybody else.

I just look at it from the following point of view: this signal line cannot be generated and cannot be processed with circuit that has digital output/input. It requires digital to analog conversion and analog signal processing for generation. And it requires analog signal processing and analog to digital conversion before it can be processed as digital.

Since it cannot be generated and cannot be processed with digital circuit and requires analog signal processing with analog circuits, it is absurd to name it as digital. Isn't it? That's my point of view.

There are also asynchronous digital signals (e.g. RS-232).

RS232 can be generated and processed with digital GPIU with a proper logic-level shifters (voltage level translators).

But 1000BASE-T signal is much more complicated than RS232 and cannot be generated and processed with digital GPIO even with logic-level shifters. 1000BASE-T signal requires mandatory analog signal processing with analog circuits (analog amplification/attenuation, analog add/subtraction, common mode current rejection, echo cancellation, etc) and it requires analog-to-digital and digital-to-analog conversion before it can be processed in a digital domain. So, attempt to use analogy between RS232 and 1000BASE-T is incorrect.

[radiolistener]

There are binary digital signals, and there are also digital signals with more levels. One does not exclude the other.

I understand your point of view, but cannot accept it as a common meaning for the term "digital".

Can you please show me the example of a clean digital circuit (not hybrid analog+digital) which uses more than two voltage levels on a digital signal line?

For example, here is the explanation of the term "digital" from dictionary:
https://dictionary.cambridge.org/dictionary/english/digital

digital
adjective

recording or storing information as a series of the numbers 1 and 0, to show that a signal is present or absent:
digital data

IT, COMMUNICATIONS
digital
adjective

using a system that can be used by a computer and other electronic equipment, in which information is sent and received in electronic form as a series of the numbers 1 and 0:
- digital content/data/information
- digital camera/photography/image
- digital television/TV/radio
- After the switchover all our TV channels went digital.

Compare
analogue

https://dictionary.cambridge.org/dictionary/english/analogue

IT, COMMUNICATIONS
analogue
adjective (US usually analog)

using a system in which information, such as sound or images, is stored or sent in a continuously changing form, such as electrical signals, radio waves, or film:
- analogue TV/radio/phone
- analogue system/signal/service
- Viewers would be forced to make the switch from analogue to digital.

[ve7xen]

1) Can you categorize output from DAC? This line (named as "analog") convey digital data which is sent on input of DAC in binary form and discrete in time.

How the signal is generated is irrelevant. For our purposes (we are not operating in the quantum domain), all signals are continuous both in time and voltage/current/power, regardless of how they are produced. Making a distinction between a 1-bit DAC (a 'digital' output) and a 5-level DAC is a distinction without a difference. They both convert digital data to a continuous analog value on the line.

2) Can you categorize output of ADC? These lines represented in binary form and discrete in time convey analog data which is sent on analog input of ADC.

I'm not sure what you're asking, of course the output of an ADC is digital. A sampled analog signal is digital data, not analog data, that is why you need to contend with quantization and aliasing that are not present in the analog domain.

Regarding to 1000BASE-T, it's signal is produced with DAC, analog amplifier with programmable gain and passed through transformer like usual analog line. This line cannot be processed with digital input. It needs analog signal pre-processing and then needs analog to digital transformation before it can be processed in digital domain. It transfers continuous analog signal which convey both - RX and TX digital data simultaneously. And of course it is processed with analog circuits such as amplifiers. And this is why I can't accept "digital" type for that signal line.

Please define a 'digital' input in a way that doesn't invoke continuous analog properties of the signal (voltage, current, etc). A 'digital' input is just a 1-bit ADC with some hysteresis, and you absolutely can use it to extract multiple signal levels from the input with sigma-delta techniques and oversampling. We don't actually generally know how a 1000base-T PHY is implemented, since it's not relevant as long as it meets the spec, but I'd wager that sigma-delta and DSP is used much more heavily than analog signal processing. Does it mean that it is a 'digital' signal to you if it is using a 1-bit sigma-delta ADC and DSP, but an 'analog' signal if it's using a 17-level ADC? What if one link partner is using a '17-level DAC' and the other is using a sigma-delta DAC?!

[radiolistener]

A 'digital' input is just a 1-bit ADC with some hysteresis, and you absolutely can use it to extract multiple signal levels from the input with sigma-delta techniques and oversampling.

No. ADC is a hybrid analog+digital device, it has analog input and digital output.

And you cannot do analog-to-digital conversion just by putting analog signal on a digital GPIO and by using high speed oversampling.

1-bit ADC requires mandatory feedback which needs analog operations (analog addition and analog integrator) with the input analog signal and it requires analog comparator before digital input, so this is a hybrid circuit which uses both - analog and digital circuits.

[langwadt]

A 'digital' input is just a 1-bit ADC with some hysteresis, and you absolutely can use it to extract multiple signal levels from the input with sigma-delta techniques and oversampling.

No. ADC is a hybrid analog+digital device, it has analog input and digital output.

And you cannot do analog-to-digital conversion just by putting analog signal on a digital GPIO and by using high speed oversampling. 1-bit ADC requires feedback which needs analog operation with the input signal and analog comparator before digital input, so this is a hybrid circuit which uses both - analog and digital circuits.

you can if you have enough noise, quite a few GPS receivers do/did that

[ve7xen]

No. ADC is a hybrid analog+digital device, it has analog input and digital output.

And you cannot do analog-to-digital conversion just by putting analog signal on a digital GPIO and by using high speed oversampling.

1-bit ADC requires mandatory feedback which needs analog operations (analog addition and analog integrator) with the input analog signal and it requires analog comparator before digital input, so this is a hybrid circuit which uses both - analog and digital circuits.

If you do not allow analog components in the signal chain, then you are operating in a purely theoretical space that doesn't exist in the real world. All signals are analog. All electronic components are analog. 'Digital' only exists as far as we have defined line codes to give meaning to particular analog values. Whether those line codes are binary or multi-value or complicated modulations isn't really relevant. All signals are analog, ergo the only meaningful distinction between analog and digital is related to the meaning / interpretation of the signal. That you are unable to see this is kind of mind blowing.

[radiolistener]

If you do not allow analog components in the signal chain, then you are operating in a purely theoretical space

No, I'm allow analog components in the signal chain. But if a signal chain has analog and digital components, then this is a hybrid circuit, not digital. Hybrid circuit can combine digital and analog components.

You can't name some analog signal line as digital just because this signal line is used in a hybrid circuit which consists of digital components. The same you can't name some analog signal line as digital if that signal line is used to transfer digital data over analog signal.

[ve7xen]

If you do not allow analog components in the signal chain, then you are operating in a purely theoretical space

No, I'm allow analog components in the signal chain. But if a signal chain has analog component, then this is a hybrid circuit, not digital. Hybrid circuit can combine digital and analog components.

All components are analog. All signals are analog. What is a 'digital' component, and how does it interact with the continuous real world? Is an inverter with feedback used as an amplifier digital or analog? Why? How about in a ring oscillator? How do digital signal thresholds work, and how is this not an 'analog' circuit, since it involves 'comparing' analog voltages?

Your view is simply not compatible with real life. Either all signals are analog (because they are), or the meaningful distinction rests on what is being represented. There is no alternative that makes sense.

[radiolistener]

All components are analog.

No, there is also digital components. They are designed to work with digital signals represented with discrete digits. The digital signal is not continuous, it has two discrete voltage levels to represent digits 0 and 1. Also digital signal is discrete in time, it should have valid and unchanged state at specific time interval determined by the the master clock signal edges. Outside of that time interval the state of the digital signal doesn't matter and can be undefined.

[radiolistener]

Your view is simply not compatible with real life. Either all signals are analog (because they are), or the meaningful distinction rests on what is being represented. There is no alternative that makes sense.

No, my point of view is from a real life. It is not theoretical, but practical.

I think you're just don't understand the key difference between digital and analog signals, and this is why you're confuse them.

You can transfer digital data through analog line, but that analog line still remains analog.
The same you can transfer analog data through digital line, but that digital line still remains digital.

It doesn't matter what you transfer through signal line. The difference here is how data is represented on the line.

The digital line represents data as discrete digits 0 and 1 separated in time with a clock edges. You can capture analog data at discrete points, encode its voltage into binary form and transfer it through digital line as series of 0 and 1. In order to restore analog data you're needs to decode series of 0 and 1 into voltage levels at discrete points and then interpolate amplitude between these points. Digital signal carries limited amount of information.

Analog line has a different data representation. It has continuous wave, and each point of that wave carries information, there is no discrete digits and no discrete points in analog signal. Analog signal carries infinite amount of information.

[helius]

There is no such thing as "infinite amount of information" since information is entropy and entropy is a finite quantity determined by chemistry and physics. In any analog signal transmission there is inevitably noise and distortion and these limit the amount of information the signal contains.

You also cannot really "transfer analog data through digital line", since as you said, any information must be encoded as bits to be transferred, and at that point it ceases to be analog. In a similar manner, we can see that digital data must be encoded as waves to be transferred over analog lines: the shape of those waves can be selected based on the fidelity of the channel, which is called the coding efficiency.

This was all worked out by Claude Shannon 70 years ago.

People forget what the "analog" term really means. An analogy is a concept that has a similar structure to another concept, to aid in understanding how that structure works. (Before it was banned for PC reasons, the section of the Scholastic Aptitude Test that was most correlated with IQ was the analogies.) An analog variable has a structure similar to a physical quantity. For example, the structure of an analog audio signal is similar to the physical displacement of a microphone diaphragm.

In modern practice, many of the interconnections in use today are no longer strictly digital. For example, the PCIe and DDR interconnects require SERDES blocks at either end, which are used to encode bits into packets of self-clocked data and condition their analog waveshape. Increasingly, digital channels at the board level are going away.

[ve7xen]

No, there is also digital components. They are designed to work with digital signals represented with discrete digits. The digital signal is not continuous, it has two discrete voltage levels to represent digits 0 and 1. Also digital signal is discrete in time, it should have valid and unchanged state at specific time interval determined by the the master clock signal edges. Outside of that time interval the state of the digital signal doesn't matter and can be undefined.

Please explain how logic thresholds are not 'analog' in your view of the world, and how adding a third (or 2048th) threshold changes the fundamental nature of the signal. How do you explain voltage drop and fanout. If I connect TTL logic to CMOS logic, is it still digital, despite them not agreeing on the 'discrete' voltages? If I use an ADC to sample a 'digital' signal for some reason, and decode it in DSP, has it become analog?

Is an NRZ-I signal digital to you? How about a Manchester code? Neither seem to fit your prior definition. Does your definition allow line coding at all, like 8b10b? How about a differential signal that uses two physical lines? What about some of the other goofy stuff that's done, like using pulse lengths to encode digital data?

i don't think I am the one that is confused here...You do realize that both time and any signal value are continuous in reality, and that until you get to quantum effects, nothing is discrete, right?