Comment by chowells | Hacker News Reader

Comment by chowells | Hacker News Reader

Parent Full thread

Wow. There's some truth in there, but a lot of nonsense. Especially in the analogies. If you can't see the difference between 60Hz and 144Hz when a small object (mouse pointer) is moving quickly against a static background, you're not paying attention. And 144Hz is amazingly far short of the actual perceptual limit in that case. Consider the difference between what it looks like to move your fingertip around and move a mouse pointer around at high speed. I haven't done tests to measure, but I'd be surprised if I mistook a 500Hz sampling rate for continuous movement.

This is all because there are edge cases for which your brain can make use of significantly more information than it can directly identify. Just because a pixel has a smaller arc than your retina can distinguish doesn't mean that its contribution to the image isn't recognized. "Retina" screens are still far below the level at which there would stop being advantages to increasing resolution. (Yes, there would be huge disadvantages too, and I'm not saying we need higher resolutions.)

What are the odds that there are edge cases where the extra bandwidth of 5G would actually matter? Seems pretty close to 100% to me. Some of those edge cases might be as easy to find as moving your mouse pointer rapidly. Maybe going to a sporting event?

The killer feature isn't throughput. Much like 2G, 3G, and 4G before it, the killer feature of 5G will be the number of low-bandwidth devices that can be connected at extremely low cost.

Ever notice how almost all home security networks support cellular connections these days, even the lowest priced options that cost $10/month? That's 2G and 3G at work. The marginal cost to the security company is probably $0. Likewise with the Kindle--so cheap that you only pay a single upfront cost premium, largely for the hardware.

Cringley mentions the desire to supplant POTS networks, but that doesn't do justice to the market potential. Home security networks switching from POTS to cellular was an incremental improvement. The growth will come from all the various devices and services that previously weren't economically viable given installation and recurring connectivity costs. Basically, IoT, except without context IoT is a meaningless phrase.

For example, I noticed yesterday that the arcade at the movie theater switched to electronic tokens. The card dispenser had a cellular antenna, thus I assume the whole thing was completely remotely managed and probably just took a few hours to install without having to provision any local services (like a POTS line) except electricity. Vending machines, etc, are obvious [existing] growth markets, but who knows where it will go.

Anybody know who's the leading provider in this space? I've always assumed Sprint with T-Mobile in second place.

>I noticed yesterday that the arcade at the movie theater switched to electronic tokens. The card dispenser had a cellular antenna, thus I assume the whole thing was completely remotely managed and probably just took a few hours to install without having to provision any local services (like a POTS line) except electricity.

I used to work at a firm that made/distributed/maintained gaming machines, and that's exactly correct. It came about in a roudabout way, though.

Some jurisdictions mandate active management and reporting from 'fixed-odds betting terminals' and historically that was done with POTS and overnight modem dialing. That moved across to 3G because the on-site complexity was drastically lower, but it was a slow move due to the carrier and new hardware costs.

Now those costs are barely anything, and the back-end hardware is already in-place for FOBT. So it makes a lot of sense to move all the other terminals over to this system even if it's only for maintenance reporting/alerting, and not odds/configuration management.

The electricity metering industry tells a nearly identical story. Now every meter might as well be a smart meter, because everything needed to achieve that has to be present anyway, it's just a case of rolling out the end customer devices.

> the killer feature of 5G will be the number of low-bandwidth devices that can be connected at extremely low cost.

We already have this. It's called NB-IoT (Narroowband IoT) and the carriers have been dragging their feet horribly.

It's chicken and egg. Without a cheap, high volume, low power chipsets, user devices don't exist. Without lots of user devices, the carriers feel no pull to roll it out. Without rollout, nobody feels the need to jump in and create a cheap, high volume, low power chipset.

T-Mobile USA has had NB-IoT commercially available for a year: https://iot.t-mobile.com/

EDIT: Press release from the original announcement, https://www.t-mobile.com/news/narrowband-iot

Nordic has the nrf91, which should be fairly high volume and low cost. I'm sure other companies have similar solutions in the works (TI and Silabs probably have something).

I do wonder what the limit is in visual “framerate”. Normal light seems very smooth, but I can easily pick up on individual “frames” if I wave my finger under an older fluorescent light (which should be running at 50-60Hz to match mains power if I understand it correctly).

The eyes don't have a framerate. A video at 60fps looks smooth but a single black frame on a white video is visible even at much higher fps. If I move my mouse quickly in circles on a 60fpx screen it looks like I have 10 individual cursors in a ring but if I move my finger quickly in circles it looks like I have a slightly transparent ring shaped finger with no gaps in between, I can see the smooth curve of the path that my finger took which would have to be captured at an incredibly high framerate to display on a monitor.

A bit off topic but I've been thinking there has to be a better way of handling this at the LCD controller level here...

Currently, a 60hz screen isn't perfect and pixel overdrive will be used to reduce motion blur. In effect, there are more 'unique intermediate frames' being displayed as the system tries to approximate the discrete input frames physically. The actual blur on fast-moving content on a modern screen is huge...

Ignoring latency for a moment, if a controller has the previous and next frame plus motion vectors and knowledge of the precise pixel response and decay curves for the monitor... it should be possible to continuously adjust the current/voltages(?) to every pixel to better approximate what a much higher refresh rate low motion blur panel would look like on high fps content. Can't remove blur but can make motion continuous rather than a stair step. A kind of smart motion interpolation + overdrive implementation.

The assumption is the higher fps the more linear the motion is frame to frame. 30 to 60fps interpolation should have more noticeable abnormalities than 120 to 1000hz.

I'd like to see a demo of this...

I doubt one could make it work for computer graphics (since motion vectors aren't easily accessible), but when displaying video the results could be good, although users seem to dislike content with interpolated generated frames, and perhaps your proposed tech could get a similar response.

Arguably computer graphics are the best candidate since many engines already have a velocity buffer for effects like temporal anti-aliasing and motion blur... would need API integration though.

My hunch is motion interpolation is preferred at high fps. E.g. a high action scene at 30 fps will need a lot more information to be filled in and has likely already been mastered with the target frame rate in mind (camera shutter speed etc).

But 120hz or more content will have very little motion between frames, and this motion will be far more linear (e.g. a finger might move 1cm between frames, but doesn't bend so less motion artefacts).

So 120hz content upscaled to 960hz+ would be much closer to native 960hz content and preferable to displaying at the original 120hz. A logical implementation would be VR headsets with a built-in DSP.

https://www.svp-team.com/wiki/Main_Page This came to mind

The individual pixels are only row-column addressable, unfortunately.

I have an idea to test, but I don't have the kind of electronics background to construct the necessary rig affordably. Or possibly to identify it if one exists already.

Here's my thought: connect a laser to a high frequency switch, similar to what you'd find in a PWM controller. Then rig the laser to move such that the beam traces a small circle relatively rapidly. Set up in a dark room, and play with the switching speed until you can no longer distinguish individual spots/blurs as it traces the pattern on a wall. Based in my experience with a 144Hz monitor, I'm betting that if it was moving fast enough, I'd still be able to tell the difference at 500Hz, anyway. Possibly a lot higher. I'd be fascinated to get empirical results back. Especially if they proved me wrong.

You could probably easily build a rig like that with a laser pointer and two servos, you don't need an extensive electronics background. The sample code put there would get you pretty far.

Servos that can move something at 500Hz are hard to come by...

Galvanometers are the normal way to do that, but they are quite expensive and not a normal part of the hobby electronics repertoire.

You don't need it to move at 500 Hz, you can even just have it spinning around with constant speed. All you need is for it to be sufficiently fast so you can check if you notice any gaps.

Gears?

There need not be any limit. The eye can pick up a of a single photon in the right context. It doesn't get any briefer than that.

That is irrelevant here, the point is how long does it take to distinguish two consecutive inputs as distinct.

Yeah, I won't believe everything on cringely .... he writes about his opinions which are often short sighted ... he does have his moments though ...

Reply on news.ycombinator.com