aDfbrtVt

Depending on phase matching criteria of lambda's on a given core, I would mostly agree that various wavelengths are not significantly coupled. I also agree there are a different power budget for LH modems vs. DCI, but power on LH modems is not something that often gets publicly disclosed. I am not too concerned with the overall power, more the power density (and component density) that 19 channel MIMO would require.

The main point I was trying to make is the impracticality of MIMO SDM. The topic has been discussed to death (see the endless papers from Nokia) and has yet to be deployed because the spatial gain is never worth the real world implementation issues.

To get a ballpark power usage, we can look at comparable (for some definition thereof) commercial offerings. Take a public datasheet from Arista[1], they quote 16W typical for a 400Gbps module for 120km of reach. You would need 2500 modems at 16W (38kW) jointly decoding (i.e. very close together) to process this data rate. GPU compute has really pushed the boundaries on thermal management, but this would be far more thermally dense.

[1] https://www.arista.com/assets/data/pdf/Datasheets/400ZR_DCI_...

As others have mentioned, this is mostly a proof of concept for a high core count weakly-coupled fibre from Sumitomo. I also want to highlight the use of a 19 channels MIMO receiver structure which is completely impractical. The linked article also fails to mention a figure for MIMO gain.

I can think of a number of cases where using a hard decision decoder can be a better choice. Power would be one factor and I would strongly disagree that the power delta between soft and hard codes at these data rates is small. Unfortunately, I can not find any public data to back up that claim for what appear (based on coding gain and overhead) an HD-FEC using RS and a SD-FEC using braided BCH or LDPC.

Other factors can include reduced routing complexity and area requirements of HD since you have to shuffle around soft information. Extra die space is expensive so you want to avoid it if possible.

However, I think the most likely is the latency reduction you get when using HD-FEC. I know that some applications of microwave links are extremely latency sensitive, could be that this research is targeting one of those applications.

Something not totally clear from the title, but it seems the claimed rate was actually achieve with a transmitter structure similar to what you would find in coherent optics (see figure 4). Instead of coupling to fiber, they couple to a high speed photodiode that radiates at the ~140GHz laser wavelength.

EDIT: Noticed that after a closer reading of the paper, the real goal was to assess the LO phase noise improvement when moving from a RF synthesizer to a SBS laser and PD based LO.

FYI, they recently removed the annual RRSP admin fee. https://www.interactivebrokers.ca/en/index.php?f=rsp_tfsa_in...

Quick correction - not repeaters but amplifiers. The term repeater is commonly used to refer to a device that demodulates and remodulates a signal.

Why so many taps? That's north of 2 seconds.

What is "reliable" though? All digital systems will undergo some rate of correctable/uncorrectable errors, it just depends on what you consider to be an acceptable rate of failure for your device.

That may hold for a trivial device or a perfectly spec compliant device. However, the former is not interesting and the later does not exist. I agree that more test coverage would be beneficial, but I think your heavily downplaying the difficulty of writing realistic mock hardware.

I fail to see how you could "disrupt" something like cooling solutions in the way I believe your post implies.

First, the ability to remove large volumes of heat from a DC (or analogous plant) is largely dependent on the local geography/environment. In modest cases, that may come down to average ambient temperatures and the cost of electricity. In more extreme cases, access to a massive heat sink such as a body of water may be necessary.

Second, the technology used to evacuate heat is very mature. The modern world depends on HVAC and has for a very long time. While there are incremental advances such as new refrigerants and compressor technology, there is always a cost to performance tradeoff.

Note that the theoretical limit on cooling performance (CoP) of a Carnot engine is ~7.8. Check out this report by the IEA (page 43) https://iea.blob.core.windows.net/assets/0bb45525-277f-4c9c-...

I would argue the FFT algorithm is much more important in communication systems than the Viterbi algorithm. There are plenty of powerful Forward Error Correction schemes that do not use Convolutional codes (and thus cannot use Viterbi algorithms). However, every single system with a reasonably long channel equalizer makes use of a real time FFT algorithm.

Fun fact, Golay codes are the only perfect code outside of Hamming codes. Perfect codes are interesting in that the Hamming bound becomes an equality, effectively saying that the distance between all valid codewords is exactly the minimum distance "d".

Looks like Intel 4nm will only start shipping 2023H2. I'm not sure I see Intel 2nm shipping less than 6 months later, especially given Intel's recent history of stumbling on execution. How firm do you believe the 2024H2 delivery date of N2 to be?

I mean yes and no? Kalman filters fuses information about known system behaviour along with observations in an optimal predictor (for certain types of noise). It uses a matrix in the same way that almost every operation on a computer uses a matrix.

Edit: I missed the quip about the Kalman comment being in jest, my reply probably wasn’t necessary but I’ll leave it up.

I have had this page bookmarked for years, hopefully you find it as useful as I do. It is a formal yet digestible treatment of floating point numbers.

https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.h...

I see, the reduction of antenna gain with terminal miniaturization makes sense. I didn't know if the noise figure of a GaN amplifier was worse.

Why would a change in the antenna radiation pattern limit the power spectral density, is it a regulatory limit?

What aspect of migrating to GaN PA's degrades the link budget for this application?

To further reinforce the above argument, many environments dominated by Matlab have their instrument control software written in Python. Mathworks have made it pretty easy to call out to Python (although no Conda/Venv support...), they definitely feel the pressure in that area.

(Double checked your username, seems we've discussed the Matlab/Python space before!)

Picard has worked really well for me.

https://picard.musicbrainz.org/

Very likely there will be a 1.6T DR4 module in QSPFP-DD/OSFP form factor before 2030 that targets data center interconnects (< 120km). The technology (mostly) exists at this point, though the thermal density and die area pose some challenges.

I don't see much partisanship in this comment section, mostly people discussing the O&G amortization schedules. I agree that, in general, it would be better if HN had less political posts. That said, just upvote the type of posts you think are interesting and move on.

To echo your last point: If we spent half as much time upvoting as individuals instead of finger pointing, we wouldn't need HN moderator intervention at all.

Add the following lines to your sway config. It works for me when using Waybar.

exec_always --no-startup-id blueman-applet

exec_always --no-startup-id nm-applet --indicator

I would not wish Matlab GUI programming on my worst enemy. I disagree with your assertion that documentation for Python is typically much better. Take a randomly chosen function I regularly use: "pwelch"

https://docs.scipy.org/doc/scipy/reference/generated/scipy.s...

https://www.mathworks.com/help/signal/ref/pwelch.html

Maybe my bias is showing, but I find the Matlab documentation to be superior. However, that is nothing compared the quality of the IDE which is critical when doing scientific compute. When I was re-evaluating our simulation environment at work, we looked at PyCharm. The plotting capabilities alone made it easy to stick with Matlab, even at the orders of magnitude higher cost.

N.B. This is my experience in an Optical Communications R&D environment.

I agree with you for the most part, I write this while running Matlab simulations in the background. Having tried to reach similar levels of productivity with Julia or Python, I find the FOSS environments always fall short in the tooling environment. Not to mention the importance of documentation.

However, I can't help but take exception to Vivado and SOLIDWORKS being good software... While powerful, they are both incredibly buggy and prone to crashing if you so much as look at them wrong.

I would recommend any of the well reviewed units from the "Tiny Mini Micro" series from Serve the Home. https://www.servethehome.com/introducing-project-tinyminimic...

I personally use a Lenovo Tiny M920q purchased used from eBay as a Proxmox node.

Why would Qualcomm design their silicon for an 8nm Samsung process? They have a working relationship with TSMC and definitely have enough volume to access risk production nodes.

Interesting, the checksum looks like a standard LFSR but uses addition instead of XOR.

1) Seed checkum with initial_value

2) Add a data uint sized to the checksum

3) Shift up by 1 bit while feeding back the MSB to the LSB.

4) GOTO 2 until you have consume all the data.

Not sure why you would process with usual addition instead of GF2 addition.

I only see references to a BCH2 ECC in ec.c source file. I would have thought a non-binary code like RS would be better suited to filesystem coding since a fault is likely to corrupt more than a single contiguous bit. I don't have any experience with FS ECC, is there a reason BCH2 is desirable for bcachefs?

Not harder, just different (modulo your definition of old school circuit design).