It's very awesome that you are doing these series of short videos. It's really convenient to have these "small bits" of important information that can be watched in less than 15 mins. Thanks a lot!
High speed is a really a matter of L* di/dt . You would have risetimes that are low and high currents to be switched you have to make some tweaks to your design. If you have RF + High speed you have to master stackup of the board. Or use a recommended stackup from the manufacturer. I am glad you mentioned the thre key things - knowledge of terminations.- thevenin or series, board stackup and how to do it properly , how to use GND planes and copper fill. But its common wisdom and backup by reliable textbooks : BW approximates = 0.35/trise. Now you did mention Fourier coefficients after about the 7th or 9th coefficient it would be almost too small to make a contribution to the Power spectrum. Even if you use a Square wave the coefficient magnitude drops off as 1/N where N is the order of the term.
High speed as we understand it only cares about the rise time because that's what determines the power spectrum of a signal. dV/dt and dI/dt matter for EMI/radiated emissions and crosstalk, but even with crosstalk we quantify signals injected into a victim line via crosstalk as a ratio, not as absolute value. But for the inherent characteristics of signal propagation, rise time is what matters.
Hi barronisme, those approximations are based on how an oscilloscope reads out a wideband signal. The front end of an oscilloscope acts like a higher order filter, meaning you can't accurately capture a rise time that is too fast. Then there is the channel's bandwidth, which is limited by multiple factors like load capacitance, dispersion, and roughness...
@@Zachariah-Peterson What I'm trying to determine is - if I have a Tr rise time, upto which frequency should I include in my analysis? I need to stop somewhere right.
@@barronisme I understand, it depends what you're trying to do. If you want to measure a signal with a known edge rate using an oscilloscope, you would want to measure at the output of the driver and use the rise time to determine the required bandwidth. If you're designing a high speed transmission line and you want to make sure it can support a required data rate or bandwidth, you'll care more about the load capacitance and any effects in the transmission line like roughness and dielectric dispersion. These can be captured in the transmission line's transfer function or in S-parameters. So for a given data rate and its corresponding Nyquist, you can determine the minimum required bandwidth, meaning you should not have much signal loss below that particular frequency to ensure the signal can be recovered easily. These problems with signal loss, as well as problems like jitter and skew, are the reason we use things like pre-emphasis at the driver and equalization at the receiver to aid signal recovery. If you want to use a transfer function (determined from ABCD parameters), you need to look at cases with different terminations at the source and load ends. If you assume on-die termination at the source and shunt/parallel termination at the load, then the transfer function is basically a low pass filter with some attenuation. I'll put together an article on this as I've had to derive these expressions by hand for each case. They are available in the literature (there are articles in IEEE or JPIER), but they are buried in different articles and they are never clearly communicated.
Hello Zach, I just recalled the relationship between BW and rise time from a book written by Eric Bogatin. Just like you said, if we need to have an assumption, BW = 0.35/ t_rise would be a recommended formula. Thanks.
Keeping in mind rise/fall time, Would it be best to consider (or get in the habit) of treating all boards as high-speed, even if they are not? Great information!
Yes that is a good mindset to have for digital boards, although that doesn't mean every single signal needs to hit a specific impedance target or anything like this. It's better to just approach it with best high speed design practices to ensure you prevent common noise problems in poorly designed high speed boards.
I know that many digital datasheets focus on rise time, but isn't this discussion about rise times and "high speed design" dependent on the logic level voltage? A change in state over time is meaningless without the units of change, after all. (I'm an analog guy, so I'm more accustomed to thinking in terms of slew rates than rise times.) Without even doing the math, I can sketch a graph on the back of an envelope that shows that any given rise time from 0 to 5V is a higher frequency transition than the same rise time from 0 to 3V3.
Hey Peter, that's a fair point, but it's not a "higher frequency" transition, but I get that you're thinking of this in terms of a slew rate. the reason we focus on the rise time is because it's used as an indicator of bandwidth in the digital signal, and the bandwidth will be independent of logic level. So for example, if you use a 3 dB rolloff as a measure of bandwidth, you're just making a comparison between two levels (the max signal level and the signal at -3dB rolloff). That 3 dB measurement is not an absolute measurement so it doesn't depend on signal level but it does depend on the rise time. Also people focus on the rise time because some explanations of high speed behavior are easy to understand in terms of propagation speed, so you can make comparisons involving interconnect length by using the rise time and propagation speed. However, when we look at crosstalk and radiation and power consumption, now the initial and final logic levels definitely matter. A 0 to 5 V transition at 1 ns will produce 5x as much radiation as a 0 to 1 V transition at 1 ns, as well as using comparably more power. This is one reason that large high-speed ICs with high IO counts (FPGAs?) and signaling standards are using lower voltages, there is a smaller swing between levels. Also with differential protocols, the receiver requires the two signal levels to cross each other during their transition in order to trigger a logic level change in the receiver's buffer, so the skew limit has to be defined as a fraction of the rise time but not the slew rate.
Should have started off with Fourier to show how the energy (not power(!)) in the freq. domain is related to the time domain (show increasing amount of sinusoids -> steeper curve. The 0.35/t_r BW approx. assumption is based on the single pole LPF - true for old style scopes - not newer ones. Should have clarified that the rise time and the clock-speed are only partially interdependent - but with the proliferation of high-speed logic gates, even low speed digital IO can have significant HF content.
""DIGITAL SIGNAL RECOVERY" just few days back i read that SDI video signal are recovered at reception from SDI cable by a specific circuit network.Kindly make video on that too some time in future.
It's very awesome that you are doing these series of short videos. It's really convenient to have these "small bits" of important information that can be watched in less than 15 mins. Thanks a lot!
Glad you like them!
A Video about DfT (Design for Testability) would be awesome!
Definitely need that!
For sure!
These videos are awesome Zach please keep them coming.
High speed is a really a matter of L* di/dt . You would have risetimes that are low and high currents to be switched you have to make some tweaks to your design. If you have RF + High speed you have to master stackup of the board. Or use a recommended stackup from the manufacturer. I am glad you mentioned the thre key things - knowledge of terminations.- thevenin or series, board stackup and how to do it properly , how to use GND planes and copper fill. But its common wisdom and backup by reliable textbooks : BW approximates = 0.35/trise.
Now you did mention Fourier coefficients after about the 7th or 9th coefficient it would be almost too small to make a contribution to the Power spectrum.
Even if you use a Square wave the coefficient magnitude drops off as 1/N where N is the order of the term.
High speed as we understand it only cares about the rise time because that's what determines the power spectrum of a signal. dV/dt and dI/dt matter for EMI/radiated emissions and crosstalk, but even with crosstalk we quantify signals injected into a victim line via crosstalk as a ratio, not as absolute value. But for the inherent characteristics of signal propagation, rise time is what matters.
I like what Dr. Howard Johnson says - if Tr is your rise time, then for most SI purposes, we can consider the signal BW upto 0.5/Tr.
Hi barronisme, those approximations are based on how an oscilloscope reads out a wideband signal. The front end of an oscilloscope acts like a higher order filter, meaning you can't accurately capture a rise time that is too fast. Then there is the channel's bandwidth, which is limited by multiple factors like load capacitance, dispersion, and roughness...
@@Zachariah-Peterson What I'm trying to determine is - if I have a Tr rise time, upto which frequency should I include in my analysis? I need to stop somewhere right.
@@barronisme I understand, it depends what you're trying to do. If you want to measure a signal with a known edge rate using an oscilloscope, you would want to measure at the output of the driver and use the rise time to determine the required bandwidth.
If you're designing a high speed transmission line and you want to make sure it can support a required data rate or bandwidth, you'll care more about the load capacitance and any effects in the transmission line like roughness and dielectric dispersion. These can be captured in the transmission line's transfer function or in S-parameters. So for a given data rate and its corresponding Nyquist, you can determine the minimum required bandwidth, meaning you should not have much signal loss below that particular frequency to ensure the signal can be recovered easily. These problems with signal loss, as well as problems like jitter and skew, are the reason we use things like pre-emphasis at the driver and equalization at the receiver to aid signal recovery.
If you want to use a transfer function (determined from ABCD parameters), you need to look at cases with different terminations at the source and load ends. If you assume on-die termination at the source and shunt/parallel termination at the load, then the transfer function is basically a low pass filter with some attenuation. I'll put together an article on this as I've had to derive these expressions by hand for each case. They are available in the literature (there are articles in IEEE or JPIER), but they are buried in different articles and they are never clearly communicated.
Hello Zach, I just recalled the relationship between BW and rise time from a book written by Eric Bogatin. Just like you said, if we need to have an assumption, BW = 0.35/ t_rise would be a recommended formula. Thanks.
Keeping in mind rise/fall time, Would it be best to consider (or get in the habit) of treating all boards as high-speed, even if they are not? Great information!
Yes that is a good mindset to have for digital boards, although that doesn't mean every single signal needs to hit a specific impedance target or anything like this. It's better to just approach it with best high speed design practices to ensure you prevent common noise problems in poorly designed high speed boards.
Thank you, Zach, for all your explanations and lessons. They are really helpful :)
Glad you like them!
5:11 the clock period you mentioned is actually only half of the clock period.
Amazingly Explained!!
Glad you think so!
I have a 1PPS distribution so 1HZ... but the rise time is ~300ps
At least trying to get to 300ps. FR4 and tuned traces I get it down under 1ns...
Awesome video 🙏.
Thanks a lot.
Thanks a lot for the highly informative videos you are providing. Please make one or more videos on RF pcb design.
Hi Zach, thank You for this video. Would be possible from Your side to make a video about crosstalk on a PCB board, if yes than thank you :)
Hi Patryk, We have been getting a lot of questions about different design topics, but I'll do my best to work it into the schedule!
I know that many digital datasheets focus on rise time, but isn't this discussion about rise times and "high speed design" dependent on the logic level voltage? A change in state over time is meaningless without the units of change, after all. (I'm an analog guy, so I'm more accustomed to thinking in terms of slew rates than rise times.) Without even doing the math, I can sketch a graph on the back of an envelope that shows that any given rise time from 0 to 5V is a higher frequency transition than the same rise time from 0 to 3V3.
Hey Peter, that's a fair point, but it's not a "higher frequency" transition, but I get that you're thinking of this in terms of a slew rate. the reason we focus on the rise time is because it's used as an indicator of bandwidth in the digital signal, and the bandwidth will be independent of logic level. So for example, if you use a 3 dB rolloff as a measure of bandwidth, you're just making a comparison between two levels (the max signal level and the signal at -3dB rolloff). That 3 dB measurement is not an absolute measurement so it doesn't depend on signal level but it does depend on the rise time. Also people focus on the rise time because some explanations of high speed behavior are easy to understand in terms of propagation speed, so you can make comparisons involving interconnect length by using the rise time and propagation speed.
However, when we look at crosstalk and radiation and power consumption, now the initial and final logic levels definitely matter. A 0 to 5 V transition at 1 ns will produce 5x as much radiation as a 0 to 1 V transition at 1 ns, as well as using comparably more power. This is one reason that large high-speed ICs with high IO counts (FPGAs?) and signaling standards are using lower voltages, there is a smaller swing between levels.
Also with differential protocols, the receiver requires the two signal levels to cross each other during their transition in order to trigger a logic level change in the receiver's buffer, so the skew limit has to be defined as a fraction of the rise time but not the slew rate.
Should have started off with Fourier to show how the energy (not power(!)) in the freq. domain is related to the time domain (show increasing amount of sinusoids -> steeper curve.
The 0.35/t_r BW approx. assumption is based on the single pole LPF - true for old style scopes - not newer ones.
Should have clarified that the rise time and the clock-speed are only partially interdependent - but with the proliferation of high-speed logic gates, even low speed digital IO can have significant HF content.
Zach looks like a mafia guy, Altium should find someone who looks nerdy
I assure you I'm as nerdy as they get.
""DIGITAL SIGNAL RECOVERY" just few days back i read that SDI video signal are recovered at reception from SDI cable by a specific circuit network.Kindly make video on that too some time in future.
I am giving an example of high speed low frequency signal:
xxxRST: __________________ㄇ______________
It is very unprofessional to draw a graph without naming the axes
For those who lazy to read books)