Svarichevsky Mikhail - RSS feed Svarichevsky Mikhail - RSS feed en-us Tue, 10 Jun 2006 04:00:00 GMT Sun, 24 Mar 19 13:43:44 +0000 120 10 <![CDATA[WeMacro and focus stacking for macrophotography]]>
Microscopy often requires stitching large panoramas (and sometimes - focus stacking), astrophotography - require stacking multiple long-exposure photos to mitigate tracking errors, star saturation, improve sensor noise and shot yield rate, and finally - decent macrophotography is now only possible with focus stacking. And then there are timelapses (like this 145440-frame 50-day timelapse), and any action photography where you can easily make 10'000 shots per day (that's a good time to talk about film).

I remember when I got my first macro lens Sigma 105mm F2.8 (about 12 years ago, ~2007) - I was surprised to see it's maximum aperture of F45, which severely degrades image quality due to diffraction. And this crazy aperture was needed, and sometimes it was not enough: dependence "depth of focus↔diffraction limit" was a brick wall which was not allowing to shoot small but "deep" objects with high (or even barely good) resolution.

But now there is a cure, and it's getting widely available - motorized rails for focus stacking:

226 shots, 50µm camera shift after each photo using WeMacro rail.
Stitched in Helicon Focus (Pyramid, smoothing=1). Lens is Laowa 25mm F2.8@F4

One of original photos for comparison:
Tue, 12 Mar 19 23:23:19 +0000
<![CDATA[Las Vegas airplane spotting]]>
On the photos with visible engine exhaust you can see how modern high-bypass turbofan engines operate - "dense" trail is coming only from the core part of the engine.

Someone had tailstrike:

Mon, 04 Feb 19 00:11:03 +0000
<![CDATA[Looking back at Flash/EPROM and X79 BIOS modding]]> 2.5Gib/sec (although you can rarely tell the difference vs regular SSD), size reached immense 1Tb... Now you have the ability to choose more reliable yet slightly more expensive memory (2-bit MLC vs TLC), not relying just on manufacturer's brand name and backups.

My motherboard is from 2012 (7 years???!!!), way before M.2 became a thing - hence PCI-E→M.2 adapter. And surely BIOS had no idea how to boot from NVMe. To my own surprise, I managed to bake NVMe boot driver inside BIOS, flash it - and it worked. If someone is still on AsRock X79 Extreme6 - feel free to use this modded BIOS (it is based on latest X79E6_3.10 BIOS dated april 2014 года). Dispite the age of motherboard and CPU - PCI-E 3.0 (which is still the latest version) worked perfectly and on maximum speed, one more sign that technology progress is slowing down.

Let's take a look at my history with flash/EPROM technology:

For me it all started from KC573RF2 EPROM (dated 1990), 2KiB chip holding BIOS of my first i8080-compatile computer - Orion-128. Just next to it - EPROM Mostek MK2716T (1980), oldest one I currently own. Then 1MiB EPROM NEC, which I've noticed in SEG Plaza in Shenzhen, and few largest UV-EPROM (2 and 4 MiB) i've got from ebay. I am still missing the very first one - 1702 (256 byte EPROM)...

Then my first "SSD" - used in my Orion-128 computer, holding 64 KiB of most useful programs. It allowed to avoid booting from magnetic tape on each start. Black protective tape removed for photography.

Among survived flash cards - 16MiB SD-card, still working on modern cameras (which can store whole 1 compressed frame on it). I am very sad about EyeFi : I've seen their office in silicon valley back in the days they were still operational, there could have been so many product ideas...

Not many probably have seen mini-SD cards, they were too quickly superseded by microsd. You can see recent tilt towards Samsung - V-NAND shows it's power. I am very happy that with microsd cards you can now also choise higher reliability/endurance for slightly higher cost (endurance series). Hopefully, other manufactures will follow the example.

The only places where Samsung is not dominating (at least for now) are UHS-II cards (with higher speed) and full size cards (with higher mechanical reliability):

Then - unusually complicated, one of the first fast USB3 flash drives (128Gb) and Samsung T5 1Tb USB type-C SSD. Inside it - mSATA SSD with USB converter (and again V-NAND).

Outside of the frame there is 0.5TiB Crucal M500 - last SDD still using 2D NAND flash, and my oldest SSDs: Vertex 30Gb (2pcs) and 60Gb (lost in mail) - they were the first fast SSD with on-board DRAM cache... Quite an achievement for 2009... I remember setting up a script to catch a moment their sales started at, so I probably was among the first users of Vertex SSDs in Russia.

Total volume of solid storage I am using exceeded 3Tb and it is now comparable to volume of spinning storage (26TiB). It looks like spinning storage will not be upgraded to a larger 8-12TiB disks. It is unbelievable that in the foreseeable future personal storage of 50-100Tb on SSD would not be considered insane. But our life would be perfected when economy of scale will finally make unlimited endurance non-volatile memory viable (PCM, e.t.c).]]>
Wed, 30 Jan 19 02:33:12 +0000
<![CDATA[Internals of quartz wristwatch "Luch" - and some overclocking]]> There are some things which were around for so long, that we got used to them and take them for granted. At the same time sometimes they are much more complicated to design and manufacture.

My personal favorites are quartz wristwatches and film cameras. They only became accessible to everyone because hundreds of billions of dollars in R&D expenses were spent there in the past 1-1.5 centuries.

I'll take a look at quartz wristwatch this time. I'll look inside "Luch" wristwatch which I got during my school times.

Watches itself looks simple and ascetic. Wires are mine. You'll see why at the end of the article:

On the back side we can see an IC which generates 32768 Hz signal with a quartz, and outputs 0.5Hz signal. This signal is feed to a coil with insane number of turns (2.5kOhm resistance), which actuates the mechanics.

Closer look. Gears has 200µm teeth pitch. They are probably made by a precision stamping. BTW most advanced wristwatches use more advanced gears - cut using ion etching, similar technology to MEMS manufacturing.

Surely we'll take a closer look at IC (clickable). It looks like at the bottom there is an quartz oscillator, in the middle - 16-stage divider. Snake at the top is high-value resistor.

After metal etch:

Now let's take a look at quartz: it's classical tuning-fork in metal case, sealed by glass. Electrodes are deposited on the top:

One can spot dark line at the edge of the quartz... Let's take a closer look (clickable):

Enhance 224-176:

Enhance 34-36:

It looks like frequency of the quartz was fine-tuned by Q-switched laser. Before lasers quartz crystals were tuned by slowly lapping which was very tedious job (given 0.004% frequency tolerance).

But why there are pyramids over whole area of the quartz? Quartz crystals for wristwatches are cut on XY axis. When you process crystals (etch, grind) - they sometimes show anisotropy of their properties (etch rate or ease of chipping), which in this case left these nice pyramids in Z-direction.

Ability to pick arbitrary cut axis is what made quartz so ubiquitous. Different axis show different dependency of resonant frequency versus temperature. XY-cut has 0 frequency error around 25-30°С, which gives best results for wristwatches on your hand.

By choosing more complex cut planes - you can get curve with 2 crossing of 0 over wider temperature range. This is how we got AT-cut (99% of quartz crystals are of this type) and SC-cut (they are more suitable for ovenized quartz generators as they have flat region at elevated temperatures).

Older mechanical oscillators had no such ability and were linear, without any intrinsic first order compensation (and had much more external factors to consider - remaining power reserve, vector of gravity, magnetic field).

But the world made full cycle here and returning back to mechanical oscillators - MEMS-oscillators with all their dependency of external factors and lack of first-order self-compensation. All external factors had to be calibrated and compensated. In order to reach high Q-factor MEMS oscillators had to operate in vacuum - which make them susceptible to helium leaks (you might have heard about recent issues of inability of recent iPhones to operate in 2% helium atmosphere). All this is for reduced package size (especially thickness) and unified manufacturing materials.

But quartz will always have it's use due to combination of many advantages (low CTE, piezoelectric effect, ability to do a first-order frequency compensation, low jitter).

Fun time
Now it's time to overclock the clock. At realtime they run at only 1V, 394x overclocking requires 4.8V, 507.4x - 7V and finally 582x - 10V.

I was unable to reach 600x overclocking to cover 10 minutes in 1 second, but even 582x is extremely fast for mechanical system.

All these high frequencies are resonant, and only a couple of Hz wide. If you go 2-3 Hz lower - watches will go backward:

The hardest part of this article was not etching microchip in boiling acid. It was filming watch to hear it tick. Modern world make us too used to noise. I even had to move signal generator to a different room with a really long cable to get rid of 50Hz hum of it's power supply which I never heard before.

If you liked this - you might like too.]]>
Fri, 14 Dec 18 06:46:55 +0000
<![CDATA[Activating Office 2019 without Microsoft account]]>
When releasing Office 2019 - Microsoft did another step in this direction and required to enter Microsoft account when activating your new Office 2019 purchase. If you don't enter account - activation will not finish, there is just no way to skip it in UI. Although, there is still a way to skip it via console activation... :

Run CMD.exe/Powershell with admin priveleges, go to "C:\Program Files (x86)\Microsoft Office\Office16" and run these commands:
cscript OSPP.VBS /act

Replace XXXXX-XX... with the actual key. Restart office programs - and enjoy your genuine and activated Office 2019 completely offline. I guess we cannot hope that Microsoft will realize the importance of having completely offline option of their products, so they will start (slowly at first) loosing users to older versions of Microsoft products and open-source alternatives.

PS. If you like OneNote - it's already time to start looking for open-source alternatives. Sorry Microsoft, I am not uploading my notes to you. ]]>
Fri, 16 Nov 18 07:00:26 +0000
<![CDATA[Sony a7III: Time for Full Frame 2 (and some PDAF striping/fake RAW)]]>

6 months ago I ebayed Sony A7II and thought that It will be with me for a while. But life is full of surprises - seller did not sent me few missing parts, I returned the camera... and by that time A7III was already released. There are quite a few nice features - BSI sensor with higher sensitivity, very pleasant high-ISO noise reduction, flexible autoISO settings... Some like larger battery and dual SD slot.

For me personally flexible autoISO is what gave me largest increase of good shoots yield. Also, now camera allows one to see the photos in the buffer while buffer is flushing to SD card.

After I was spoiled by zooms during ancient A-mount era ( I had Sony 16-50 F2.8 lens - it was quite sharp in the center and versatile) - here I bought loved by many Sony 24-105 F4. It was well worth it - even sharper and obviously slightly wider zoom range. Size and weight is probably maximum suitable for casual travel. F2.8 full-frame zooms in my view are only for professionals who used to lug a lot of equipment.

Fake RAW/star eater issue is still there. You can read more here, and sign a petition to Sony here. I was also able to reproduce PDAF striping issue under very exotic lightening conditions - but this is a topic for a separate article.

At the end, I really like the camera. It will definitely stay with me until 60+ megapixel full-frame cameras with 4K/60P full sensor readout are available. We might have to wait until 2022 for that...]]>
Sun, 01 Jul 18 20:12:56 +0000
<![CDATA[Soldering practice KIT - HKT002]]>
There is an 555 which frequency is set by long chains of capacitors and resistors. CD4017 johnson counter to drive 10 LEDs.
These 10 LEDs are driven by BJT's with high hFE and diodes in the base circuit, so any surface contamination from flux would not allow them to turn off LEDs.

Highly recommend to anyone who needs to solder from time to time. If you're just starting learning - you will need at least 2 set and lots of spare parts.

Here is last glitch: You can see 3 LEDs could be lit at the same time by CD4017. Why could that be?

One cannot use CMOS digital IC without decoupling. After adding decoupling it is rock solid:
Sun, 10 Jun 18 12:24:31 +0000
<![CDATA[Analog Devices AD9361 — when microelectronics is more profitalbe than drugs]]> still often an FPGA), external filters and PA if your task requires it.

Finally I was able to take a look inside and peek at manufacturing cost of a microelectronic device with such an exceptional added value.

After decapsulation we see 4336x4730 µm 65nm die. On top metal you can notice PLL's inductors and datecode - chip was somewhat ready 2 years before introduction:
Read more on →]]>
Fri, 25 May 18 07:33:02 +0000
<![CDATA[Weekend laser galvoscanner fun]]>

I exposed many cameras to direct laser beam many times in the past, but only now I got image sensor of my smartphone damaged (only color one, B&W sensor intact). That was quite puzzling at first, and probably I will need to write a short article later on on why some cameras are damaged by laser and some other (color) cameras are not.

Mon, 07 May 18 00:27:09 +0000
<![CDATA[PaperBack - proper way of storing information on paper]]> PaperBack made by Oleh Yuschuk.

While playing with it I was able to store ~500 KiB of data on a single side of A4, which could already have some practical use. This density is achieved at 300dpi data density, 80% dot scale (recommended value of 70% gave higher error rate) and 20% for ECC correction. For reliable recovery scanned image had to be slightly sharpened using Gimp2/unsharp mask, but it feels like this is the limit (ECC had to recover ~10% of errors). On 200/240dpi data density everything is much more reliable.

One can for example take a photo of the sheet using film camera and get data microfilms at home ))) Also, this data is easy to read even in distant future and does not depend on spefic reading hardware, so even aliens or humans 1000 years in the future who find a timecapsure with it would be able to read it...

Here is how data looks at 80dpi:


Now data at 300dpi, maximum for 600dpi printer:

Even closer (square side is 2.97мм). One can see that using less than 2*2 pixels for 1 bit of data would require different recovery approach due to very high rate of errors which will be pattern-dependent. Paper fibers would also cause some issues as higher data densities.

Sun, 17 Dec 17 21:19:42 +0000