lpmay

I use Anki too, but probably not as much. IMO Anki is a tool for memorizing things, not learning them. You don't make a card for anything you don't already understand.

Don't get me wrong, I think memorization is still useful for things like Haskell. You already have to have learned the concepts, but Anki helps you migrate what you've learned to a part of your brain where you just know them, no extra conscious effort required. Eventually this has to happen with what you already know if you are going to tackle bigger concepts, Anki just streamlines it.

LFSR's are cool. Jason Sachs also has a great article series on LFSR's on Embedded Related, if you're interested in an even more in-depth coverage.

https://www.embeddedrelated.com/showarticle/1065.php

There's not much of substance to this article. The world of electronics is a bigger place than turning a load full on or full off with a Mosfet. A couple exercises if you want to see why having some BJT's around is handy: Try to find a Mosfet that turns on at 0.6 Vgs. Compare the leakage Ids leakage of a Mosfet to the Ice leakage of a comparable BJT. I also take issue with the idea that "all that extra stuff" is drawn in the Mosfet symbol for "no reason". The body diode alone is an important factor to consider EVERY time you use a Mosfet. When you read schematics every day, you're trained to recognize when diodes will forward conduct in your circuit and having that drawn in the Mosfet symbol helps you immediately see problems that would otherwise make it into your final circuit.

When this is done, does it mean any LLVM language can compile to RISC-V? Or is there additional per-language work that would need to happen first?

The Nyquist rate is the theoretical minimum that allows you to distinguish one frequency from another. Sampling a higher frequency will cause an alias component to appear. This means that in practice, there must be analog band-limiting before the signal is sampled to reject any out of band signals before they are aliased to an in-band frequency and cannot be removed. The closer the band limiting cutoff comes to the nyquist frequency the more extreme the roll-off requirements for this band-limiting. Oversampling relaxes the band-limiting requirements, and can also increase the bit depth at lower sample rates. See PDM microphones for an example where this is taken to the extreme.

I think it's important that they are talking about lines not line-segments. By definition, lines extend to infinity in both directions. But I agree, the wording is still confusing. I would have worded it more like "if three lines pass through the same point and the edge of a circle, then the point is the center of the circle and the lines lie along the radii of the circle."

I don't think it is exponential. Dynamic losses in a gate arise from a couple factors (shoot through, dissipation capacitance) but as far as I know they are a joules per switching event effect which scales linearly with clock.

Oops.

I'll leave the reply as an example of some of the factual problems with the article.

Thanks!

In case anyone else is interested: the distinction between phase delay and time delay comes into play if bandwidth is large as you say, but in one more case as well. If the array is large enough that the time delay across the array is on the order of a symbol period you must use a true time delay. If you use a phase delay instead of a true time delay, energy from separate symbols will be added together to make the final bit decision and the resulting ISI will increase BER.

I think abstrakraft's comment may partially be a reaction to the informal "fast and loose" style of your article. That said, I think there are also quite a number of factual inaccuracies. I won't go into all of them, but will comment on one paragraph from the article.

"(There's another concept called "antenna efficiency" which basically says you can adjust your scoop to resonate at a particular frequency, rejecting noise outside that frequency. That definitely works - but all antennas are already designed for this. That's why you get different antennas for different frequency ranges. Nowadays, the only thing you can do by changing your antenna size is to screw up the efficiency. You won't be improving it any further. So let's ignore antenna efficiency. You need a good quality antenna, but there is not really such a thing as a "better" quality antenna these days, at least for wifi.)"

Here you are conflating the issues of bandwidth and efficiency. Efficiency is how much RF power is lost to dissipative mechanisms in the antenna compared to how much is actually radiated. What you go on to describe is the bandwidth of the antenna. Your explanation that the bandwidth of the antenna is somehow chosen to reject noise for the receiver also demonstrates a misunderstanding of how all modern receiver architectures work. The noise bandwidth is either set by an IF filter in a super heterodyne receiver, or in the case of a synchronous digital receiver, in the matched filters. Antenna bandwidth may occasionally be chosen to reject interferes, but this is a different concept entirely. I suspect in consumer wi-fi antennas, if bandwidth is designed at all beyond "enough" it would be to maximize impedance bandwidth to allow for loading. It is also patently incorrect to say that "the only thing you can do by changing your antenna size is to screw up the efficiency". Larger antenna apertures provide more focusing, and therefore more directivity, and for a given efficiency more gain. Considering an array of antennas as a single antenna is one example of this. Another would be dish or horn type antennas which can be made with larger and larger apertures for more and more gain.

There are other issues throughout the article, but it's really not my intention to go line by line. It's also not my intention to discourage you from investigating this stuff further. I'm a practicing electrical engineer, and I feel every day all I learn is how much more I have yet to learn. Suffice it to say many of these concepts are very nuanced and complex (more so than your article reflects), so I applaud you for trying to figure this out on your own. Just don't think it will be nearly as simple as learning a new programming language. A couple basic things I would recommend you look up: the difference between directivity and gain, and the difference between coherent vs non-coherent combination. If you can, find someone who really, really knows this stuff to bounce your thoughts off of. Best of luck!

I think the "phased-array radar" you're referring to would be more precisely called "active-phased-array radar". Yes, the concepts are similar. By providing a calibrated time delay between elements of the array, the direction of focusing can be controlled. I'm not sure how this is done in practice for actual radars. If it is done in hardware, the array would have a single steerable pattern. If instead each array element has a clock-synchronized direct conversion digitizer, the beamforming can be done in the digital domain and the number of logical beams would be limited by the DSP capacity available.

Depending on the context, your friend may be right.

Power dissipation in a CMOS gate is made up of two primary components: static leakage and dynamic losses. Dynamic losses are related to the "dissipation" capacitance (Cdiss or Cpd in CMOS datasheets) and the switching frequency. If you want to understand why, consider charging an RC circuit. How much energy is lost to the resistor and how much is stored in the capacitor at the end of charging? How does the selection of the resistor and capacitor value impact this ratio?

So for minimum dynamic loss you want to minimize Cdiss which involves making gates as small as possible. However, this makes the static leakage higher. So it may not be universally as simple as "a smaller process is lower power". For a system which spends most of it's time in sleep, or clocks slowly it may actually be better to eat the dynamic loss of a larger process in order to get the lower leakage, which is something I believe TI did with some of the FRAM '430 parts (but I can't find the link now).

I had a friend who worked on a very deep sea drone. He told me that it was better to flood as much of the drone as possible to minimize the strength of the pressure vessels they had to build. The one exception to this was the buoyant components, which had to be hollow to work. These were made of hollow ceramic spheres of exacting sphericity. A sphere provides maximum enclosed volume and strength for a minimum amount of structure, so was a good choice for the pressure vessel shape. They had to be so perfectly spherical to force the pressure to be balanced across the sphere and distribute the compression evenly.

Not sure if this was the exact drone he worked on, but the spheres look like what he described.

http://www.whoi.edu/oceanus/feature/floating-without-implodi...

This reminds me of Buckminster Fuller's cloud 9 concept. His idea was that since volume scales so much more strongly than surface area for a sphere, if you make a larger and larger sphere you will enclose cubed more air for only squared more structure. Build a large enough sphere and heat the enclosed air a couple degrees above ambient and the entire structure would be buoyant, like a soap bubble with human breath in it.

http://en.wikipedia.org/wiki/Cloud_Nine_%28tensegrity_sphere...

I absolutely agree about the tone. Non-CS major myself, but I don't get the impression that things like operating system design count as "ditch digging". I do think the question of whether CS is intrinsically useful in the same way as physics or engineering is an interesting one. Hope to hear a CS major's response.

>It's impossible to make things without parasitic capacitance because without a perfect insulator...

Maybe a nit, but I don't believe that is technically correct. Parasitic capacitance (and inductance) exist between any two objects because the electric (and magnetic) field extend from a charge (or current) to infinity. These fields store energy, even in perfect vacuum. The lumped circuit model of stored electric energy is capacitance, the model of stored magnetic energy is inductance so anything physical which carries charge or current will have capacitance or inductance (will store finite energy). An insulating material (dielectric) in a capacitor can increase the energy storage (by the sqrt of the dielectric constant of the material) but non-ideal materials are not necessary for the "parasitic" effect to occur.

From Tesla's website it looks like they take ~1 Hr to charge (1C rate). That would make me think they are limited by the maximum charge rate of the cells.

It would have been interesting to see how well this would work with that canonical phone that was going to have the side-scroll sensor in the side of the phone, completely off the screen.

Yup, such tools exist. I've used peat before for stuff like this:

https://github.com/sjl/peat

Take this as the kind of half-cooked thought it is at this point (long day, about to go to bed and don't want to crack open the textbook), but the fact that your path loss leaves 10dBm of power above your Rx sensitivity doesn't mean you have 10dBm of link margin. I'm not completely sure, but I'd guess it means you have 10dB of SNR (theres a units from dBm to dB thing I haven't completely resolved in my head). Whether that makes your link work or not depends on the bitrate, modulation you choose and BER you can accept (and etc). If your modulation scheme needs 9dB of SNR (proportional to Eb/N0) to operate at an acceptable BER, your system may only have 1dB of actual link margin.

Can you explain more what you mean? I think you're getting at something I agree with, but you're also saying some things I don't think are quite right...

Firstly I don't know what you mean by the Friis equation "assumes an isotropic antenna". It explicitly accounts for the gain of each antenna, in the direction of the link, relative to an isotropic radiator. The fact that it's relative to an isotropic radiator is just how all gains are measured. In fact, the antenna pattern doesn't matter at all to the link, only the gain in the direction of the link (Friis assumes no multipath). Whether the receive and transmit antenna are "the same" or "different" really doesn't make a difference.

This is the distinction I think you're trying to make, and which I agree with:

If gain is held constant, link margin improves as frequency decreases because of reduced path loss at lower frequencies. If instead antenna aperture is held constant as frequency is lowered, antenna gain will decrease at the same rate as path loss improves and there is no net effect to the link.

Both answers are technically correct, in a fixed gain scenario where your antenna can grow as large as necessary to hold gain constant, (or the scenario doesn't allow for high gain, narrow beamwidth antennas) lower frequencies will make longer links. In systems where aperture is constant (which is the case for many practical systems) antenna gain will improve as quickly as path loss degrades when you go to higher frequencies, and there is no net advantage at any frequency.