neutral density filter design

[archaeopteryx]

So, I got curious and plugged thin film properties from refractiveindex.info into OpenFilters and had a go a number of ND filter implemenations. I don't have access behind the OSA paywall, so the kinks at 400 and 465 nm are probably not realistic, and OpenFilters doesn't do combined optimization of front and backside filter coatings. As a start, here's the reflection (R, lower and upper curves) and transmission (T, center curves) resulting from an ND64 (optical density 1.8; six stop) design with eight frontside coating layers, two backside, and a straightforward set of optimizer goals (the 1.56% transmission target for an ND64 is the straight black line).

Subscribe to see EXIF info for this image (if available)

A disclaimer is I'm not enough of a materials scientist to know if all the layers are of physically realizable thicknesses and that they'll bond to each other. The stackup is based on commonly used MgF2, SiO2, and ZnSe dielectrics for spacing and antireflection with Iconel 600 alloy and Rh providing optical density. Simulation's set for BK7 optical glass and air at standard temperature and pressure above and below the filter. Layers are of plausible thickness based on the literature and a tolerance analysis based on what I know of likely sputtering accuracy produces results which seem consistent with real filters. Not surprisingly, filter properties are most sensitive to tolerances on the metal layers (the Inconel and rhodium) and the variation in optical density comes in below a third of a stop. In rough terms, given the target of ND64 most of the production yield is likely to be close to that with the worst cases probably in the ND56 to ND75 range. I'd be surprised if filter manufacturers test or cull, so this guarantee by design is probably analogous to what they do.

Subscribe to see EXIF info for this image (if available)

Subscribe to see EXIF info for this image (if available)

Photographic filters often utilize multi-path interference, typically in antireflection coatings, but also in some forms of absorptive neutral density filters such as this design. There's plenty of background available on how they work if one's curious, such this starting point, but for an ND the basic thing is to let the optimizer work out the set of materials and layer thicknesses which make its optical density (transmission) most consistent across frequency and angle of incidence. Having gone through this exercise it seems to me some of the trouble folks have reported with ND filter vignetting probably stems from the manufacturers not designing for a wide enough range of incidence angles.

For those interested in construction details, the stackups here are MgF2, SiO2, Rh, SiO2, MgF2, ZnSe, Inconel, SiO2 in front and SiO2, MgF2 in back with thicknesses between 7.5 and 90 nm. The front and rear optical index profiles are below. Rhodium's the n = 2 bump and Inconel's the n = 3.0 bit.

Subscribe to see EXIF info for this image (if available)

Subscribe to see EXIF info for this image (if available)

 

[archaeopteryx]

The approach used for the ND64 above is extensible across the most commonly used range of optical densities (one to ten stops; 0.3 to 3.0; ND2 to ND1000), resulting the progression of frontside coatings below. From front surface to glass, the stack consists of the antireflection coating (MgF2, SiO2), the attenuating metals (Rhodium and Iconel 600 alloy) and the dielectric layers between them (SiO2, MgF2, ZnSe), and the dielectric separating them from the filter glass (SiO2). The backside coatings are held constant as a control in this case, though they might well vary for optimization of production filters.

Subscribe to see EXIF info for this image (if available)

I was anticipating increasing metal thickness with increasing optical density as the metals are responsible for light absorption in NDs of the type considered here. The relatively constant antireflection coating is also expected as it's performing the same impedance matching between air and Rhodium regardless of the optical density. The increase in Rhodium-Inconel and Inconel-glass separations is a surprise and is potentially a limitation in the thin film property data that's available. A kink in the metal thicknesses found by the optimizer occurs at ND500 (nine stop; 2.7), which corresponds to a quarter wave interaction with the kinks in the Inconel 600 optical property estimate noted as limitations in the first post of the thread.

Subscribe to see EXIF info for this image (if available)

However, I find this illuminating regardless of the details. ND1000s are known for being less neutral and having more variability in transmission as a function of wavelength than less dense designs. I've long taken this as a signal high optical density filters are harder to make but wasn't clear as to why since thicker coatings tend to be less sensitive to manufacturing tolerances. This analysis suggests the difficulty is likely the metal layers have become thick enough to have wavelength specific interactions with visible light rather than being consistently behaved across visible wavelengths (equivalently stated, higher optical densities shift such interactions out of the ultraviolet, which photographers are mostly unconcerned with, and into the visible where they're a more---ah---visible problem).

I suspect more costly filter implementations can circumvent this limitation. But a testable prediction here is that, if one's seeking to maximize neutrality with low to moderate cost ND filters stacking an ND8 and an ND64 may be preferable to an ND1000. This is consistent with my anecdotal experience but may not apply to some of the most recent generation ND filters.

 

[archaeopteryx]

A logical extension of previous post's suggestions for getting round internal resonance difficulties within ND1000s by using multiple filters is to implement an ND1000 by stacking two ND32s. If these ND32s are fabricated on the front and back sides of one filter, then they can be jointly optimized and potentially outperform single sided ND1000s. As mentioned above, this isn't supported by OpenFilters (its optimizer only runs on the frontside layers) but can be sort of gotten around with a lot of manual tinkering with the backside layers in conjunction with numerous frontside optimization passes. The sequence of charts below presents comparative transmission results for

• A quasi-optimized proof of concept ND1000 integrating two approximately ND32 filters.
• An optimized single sided ND1000 with 12 layers plus a two layer backside AR coating.
• An optimized single sided ND1000 with 16 layers plus a two layer backside AR coating.

Despite being only partially optimized, the ND32 pair has somewhat less transmission variation and better NIR rejection than the single sided ND1000s and is competitive on layer count. As with the above analyses, the kinks in the charts at 400 and 455nm are limitations of the Inconel 600 model. All three implementations use comparable rhodium, chromium, and Inconel 600 metalizations.

8+8 = 16 layer combined front and back ND32s

Subscribe to see EXIF info for this image (if available)

12+2 layer ND1000

Subscribe to see EXIF info for this image (if available)

16+2 layer ND1000

Subscribe to see EXIF info for this image (if available)

Based on the above results, I would speculate recent improvements in ND1000s might indicate I'm not the only one to have noticed this. As one would expect, the layer stackups of the front ND32 and rear ND32 are somewhat mismatched as this improves cancellation of their individual variations. As a result, their optical densities and reflectivities (not shown) differ. Essentially, the design constraints are to keep them similar enough to be able to work together but different enough to maximize the improvement resulting from use of two distinct layer stacks.

 

[Petrochemist]

In general, filters other than clear protectors rely on multi-path interference and that includes absorptive neutral density filters such as this design. There's plenty of background available on how they work if one's curious, such this starting point, but for an ND the basic thing is to let the optimizer work out the set of materials and layer thicknesses which make its optical density (transmission) most consistent across frequency and angle of incidence.

I think most photographic filters will be based more on dyes embedded within the medium than done with coatings. Certainly this is the case for Cokin filters & gels such as the Kodak Wratten series. Surface coatings (as discussed in your interesting linked semrock article) are common for anti reflective use in photography but typically too expensive/complicated for more basic usages. The only photographic filters I know of that use interference for tuning the transmission are the Baader U2 (& clones), various hot mirrors & some top end ND filters that reduce NIR too (Dye based ones transmit quite a bit of NIR). Most of my filters sadly don't even have the AR coating as they tend to be older or budget versions.

I don't have any more than a fairly basic theoretical understanding of interference based filtering, but have quite a bit of experience in absorptive changes as this is often used for chemical identification and quantitation in my work. The principles used being fairly well summed up by the Beer-Lambert law, (absorbance from a species at any wavelength is proportional to it's concentration & path length).
Oddly the bits of the spectrum I use for work are almost exclusively outside the range even a 'full spectrum' modified camera will see. (With the exception of mercury specific detectors & a 400nm detector, I only really use UV below 260nm & mid IR in the 2500 to 25000nm region).

 

[archaeopteryx]

Thanks for the feedback; missed your post but I've adjusted the wording there for clarity.

I think most photographic filters will be based more on dyes embedded within the medium than done with coatings.

I'm sure that's the case for color correction or black and white contrast enhancement filters. With digital color management it's my understanding clear protectors, polarizers, and NDs dominate current sales, probably along with UV filters used as protection filters. Clears are just the AR coatings (though a few uncoated ones are still on the market) and it's my (possibly incorrect) impression UV filters are commonly implemented through glass selection (though the distinction between dying and varying the constituents of a glass to obtain a desired transmission curve is a blurry one). So far as I know the linear polarization and quarter wave plates in circular polarizers aren't dye based. By extension, variable NDs wouldn't be either.

Among fixed neutral density filters I'm aware of three main implementations:

• Dye based ones using resins, Schott's NG series, and possibly others (though Hoya doesn't seem to have competing glass). From an exchange of a couple emails with Singh-Ray quite a few years ago I suspect their process for making graduated NDs was, at least at the time, dye based and may have relied on controlled dye diffusion in the resin to produce gradation. I would suspect softer grads also tend to be dye based due to the cost of using many layers to smoothly vary optical density, though I lack a good understanding of the extent to which it might be possible to use needles or other layer tapering mechanisms to reach lower price points.
• Oxide coatings. These tend to cast brown and, from my surveys of the ND market, manufacturers seem to be moving away from them due to both the cast and declining NIR optical density similar to what you mention for dyes. Last I checked, B+W seemed to be something of an exception.
• Metallic coatings, as considered above, with Inconel 600 appearing to be the most common basis. This has come up in other ND threads here in the past---not sure if you were on them---but it appears these are increasingly common even in NDs which aren't necessarily marketed as NIR managing. Hoya PRONDs, whilst expensive, are around half of upper tier ND pricing and exhibit NIR properties consistent with this category. Recent Haidas do as well (NanoPro MC, maybe also MC II). I haven't personally tested Aurora's low cost NDs but they appear to be NIR and there's evidence suggesting at least some of Zomei's even lower cost NDs are too. NiSi is another NIR marked option around Hoya's price point.

Some hard graduated NDs, such as Aurora's variable grad, also seem unlikely to use dyes.

The principles used being fairly well summed up by the Beer-Lambert law.

In the case of metals, it's my understanding Beer-Lambert's approximation of Maxwell breaks down due to the higher conductivity (though there might be some fractal nesting of the concepts, e.g. Solis 2015). One starting point for understanding metallic ND filters is the widely used Drude model (so lots of sources besides the linked one). As a short summary, the main reason ND filters often cast blue is likely metals' plasma frequencies (where they become optically transparent) fall around 3 PHz, or roughly 100 nm. This is close enough to VIS that transmittance starts to increase.

The other major consideration is bound, rather than electron gas, interactions. Plasmonic metals (i.e. copper, gold, and silver) are challenging to use in ND filters since their resonances tend to defeat the objective of flat transmission (e.g. Reddy 2016). Less resonant metals exhibit reduced variation (in filter terms, their poles are farther off axis). A basic ND filter design operation seems to be trying to select different metals whose effects to cancel to obtain a reasonably neutral filter. This appears difficult with pure metals in individual layers, so I suspect much of metallic ND filter optimization relates to choices in thin film alloy composition. As theoretical framework for this, one's trying to suppress the Lorentz portions of a Lorentz-Drude model (for example, NNSE 508-6, Li 2017).

In particular, Inconel 600 is ~72+% nickel with around 15% chromium, 8% iron, and some cobalt. It's apparent from the individual metals' properties this is a reasonable starting point for cancellation but I haven't found any source which attempts a Drude-Sommerfeld, Drude-Lorentz, or more detailed explanation of its observed properties. My motivation for selecting rhodium is it's fairly effective in compensating Inconel variations, at least within the models I've implemented for OpenFilters based on the best optical property measurements I've been able to dig out of research library stacks. From a look at where these transition metals are in the periodic table I suspect underlying mechanisms related to d and f shell diversity and avoidance of the outer shell freedom associated with plasmons. However, I've had too much other stuff going on to have got farther than that.

I don't have any more than a fairly basic theoretical understanding of interference based filtering

Here, it's a second mechanism for seeking constant transmission. The metals in the stack provide most of the optical density but, by doing so, leaky waveguides are formed in the dielectrics between them. Give the optimizer sufficient degrees of freedom and it's able to arrange the layer set such that the overall behavior's a weak function of the angle of the incidence. Resulting ND designs exhibit minimal vignetting and color shift across the angle of view. In my experience, multiple photographic filter manufacturers' nano coatings for dust and water exhibit weak polarizing properties. I didn't consider this as those layers aren't included here but a good nanocoated ND design would probably also optimize across incident polarization.

Like many of the numerical methods I've worked with over the years, this isn't a process I'd describe as readily human comprehensible due to the complexity of the interactions. At least for the moment, however, my conceptual model of it is waveguide detuning and cancellation among the residual couplings. The electromagnetic mechanisms are different and the process appears easier than alloy selection as the waves and electrons are less tightly bound than they are in Lorentz interactions. Also significant is the behavior of the AR coating. Since transmission is low in NDs of higher optical density (say, somewhat arbitrarily, ND32ish and up), variations in reflection which are negligible at lower optical densities become important to neutrality and minimizing variations in transmission. At least with the models here it's not possible to meet flatness criteria when changing metallization without concurrent re-optimization of interference in the AR layer.

 

[barry]

And they say Rocket Science is hard!

 

[Petrochemist]

And they say Rocket Science is hard!

I've seen messages from people working at NASA that claim it's not anymore.

 

[barry]

I've seen messages from people working at NASA that claim it's not anymore.

Yeah, that's what my physics Prof said when I was in college.
This does seem more complicated.

 

[archaeopteryx]

From time to time, most recently by @Bytesmiths, I'm asked about use of ND filters for solar photography. Solar filters are often based on metal thin films. Aluminium is a popular choice (Chkhalo 2018, Chkhalo 2016, Xiu-Wei 2010, Schumacher 1977) and is used in Baader's AstroSolar film. I also considered it for the filters above but did not use it as the rhodium-Inconel 600 combination proved more neutral. I've also come across other metals in the literature including iron, tantalum, and niobium (Kong 2017). Much of solar filter design is usually devoted to selecting a narrow wavelength range of interest, such as XUV or the 656.28 nm hydrogen alpha line, and rejecting as much of the rest of the sun's light as possible to avoid instrument overheating and damage. Resulting filter designs are similar to the ones described above, just with the layer structure tuned for the narrowband purpose rather than for maximum neutrality (US7397604, US3897140).

As one might infer from the second post above, increasing the optical density of a ND filter based on thin film metals is primarily a matter of adding more and thicker metal layers. A range of effects are possible but, for infrared wavelengths, optical density is likely comparable to that in the visible wavelength (Krayer 2019, Stenzel 2019, Kaplan 2018, Hadley 2012, Hooper 2008, Kovalenko 1999). An associated question is whether, instead of buying a solar filter, can one safely stack ND filters one already has. The default answer is no. Unless the filters aren't indicated for the purpose there's no ready way of knowing if they're designed as such or if the optical density declines outside the visible band. For example, I'm aware of some ND1000s which reduce to ND2 in the infrared. Filters marketed as IR NDs are hopefully better behaved than this but very few photographic manufacturers publish transmission curves and the ones which do don't provide micrometer wavelength data. So, in general, nontrivial testing is required. Non-photographic optics manufacturers typically specify their products more completely but, realistically, a photographic ND100000 solar filter will cost less and has a good chance of being more neutral.

In certain cases, however, more definitive statements can be made. Since I designed the ND64 and ND32 + ND32 = ND1000 it's simple to indicate their behaviour when combined as an ND64 + N1000 = ND64000 stack (logarithmic addition is multiplication). Optical glasses are typically rated for use from 265-365 nm up to 2000-2500 nm but the materials data used in the filters is valid from 188 to about 2000 nm so the figure below doesn't cover the entire range of interest. It does cover, however, show the designs are at least somewhat competently IR ND with constrained variation optical density around the 0.0000156 transmission target up to the limit of model validity. It's kind of hard not to arrive at this property when designing with Inconel 600 (note the lack of optimization goals beyond 700 nm above), though probably some ND filter manufacturer has managed.

Subscribe to see EXIF info for this image (if available)

Whilst I've used the standard approximation of the sun as a 5250 K blackbody for terrestrial radiation here a more detailed irradiance model wouldn't change the basic result, just make the irradiance curves wigglier.

As something of an aside, it follows by extension from the third post above that, if one needs about 16 stops for non-solar long exposures, stacking an ND64 and ND1000 may offer more neutral results than an ND100000 as the stack has more surfaces available for metallization.

 

[Bushboy]

I have some spare glass inserts for my welding helmet...
Bytesmith could double them up.

 

[Petrochemist]

I have some spare glass inserts for my welding helmet...
Bytesmith could double them up.

Spare welding glass is very cheap, not of high optical quality but still usable despite giving a very significant green cast.
The standard 'Shade 8' is close to a 10 stop (shade 10 is roughly 13 stops & shade 13 roughly 17 stops).
Unlike many photographic ND filters welding glass always blocks UV & NIR to at least the same extent as visual light.

 

[archaeopteryx]

Unlike many photographic ND filters welding glass always blocks UV & NIR to at least the same extent as visual light.

Something I've learned about optics over the years is strong statements tend to be false. Aulektro, Makrolon, and Rhamani 2016 include glasses with higher NIR than visual transmittance or with NIR comparable to visual. Philips indicates only their shade 14 as meeting ISO12312-2:2015 solar protection requirements, contraindicating all of their other shades. ISO itself indicates much the same and also notes ISO12312-2 applies to afocal imaging, meaning shade 14 before all lenses rather than later in an optical path. NASA is somewhat more permissive, indicating shades 12 and 13 in addition to 14.

In general, my sampling of the literature suggests lighter shades are more likely to have reduced optical density in the infrared. So they may not be good choices for stacking to obtain a darker shade when infrared attenuation is important. Probably there are some lighter glasses which maintain attenuation into the infrared but I've not yet come across one with such a specification.

It's wise to be cautious of the green bump. I've seen it specified as low as ND6, though closer to ND200 is indicated as more typical in darker shades (Willmorth 2012).

 

[barry]

I picked up a Gobe ND1000 filter recently; if anyone knows if it's IR/solar-safe I'd appreciate the info.
It's marked MRC 16L, 72mm.

 

[Petrochemist]

Something I've learned about optics over the years is strong statements tend to be false. Aulektro, Makrolon, Rhamani 2016, and Moss 1979 include glasses with higher NIR than visual transmittance. Philips indicates only their shade 14 as meeting ISO12312-2:2015 solar protection requirements, contraindicating all of their other shades. ISO itself indicates much the same and also notes ISO12312-2 applies to afocal imaging, meaning shade 14 before all lenses rather than later in an optical path. NASA is somewhat more permissive, indicating shades 12 and 13 in addition to 14.

In general, my sampling of the literature suggests lighter shades are more likely to have reduced optical density in the infrared. So they may not be good choices for stacking to obtain a darker shade when infrared attenuation is important. Probably there are some lighter glasses which maintain attenuation into the infrared but I've not yet come across one with such a specification.

It's wise to be cautious of the green bump. I've seen it specified as low as ND6, though closer to ND200 is indicated as more typical in darker shades (Willmorth 2012).

The first two of your sources show spectra with the greatest transmission in the visible for all shades though NIR is close in a few.
The third one I find most strange it mentions all the filters are plastic not glass, and includes 'unnamed' filters from the Iranian central market which IMO aren't welding glass at all. The spectra being far more similar to black polycarbonate & similar materials. They specifically state it doesn't meet the ANSI Z87.1 transmission criteria.
Your fourth source refers to 'transparent welding curtains' again a very different material, designed to stop stray UV not allow viewing of the welding in process.

The Phillips reference states Shade 14 allow indefinite solar viewing. Even keen eclipse watchers are unlikely to be staring at the sun for more than an hour or two, so the more permissive shades given by NASA are still adadequate. The note from ISO only states shade 14 is suitable without giving any info on the lower shades.

All the spectra I've seen from welding suppliers as well as those I've measured myself show visible transmissions greater than UV/NIR (to 1100nm) The calculated ND equivalents will certainly be approximate Shades are determined using specific wavelengths, must be darker than the limits & the transmission is far from neutral.

The darker shades are beyond the range of the instrument at work - stray light makes results less than 0.02% transmission inaccurate even on a high grade dual monochromator spectrometer - but for the lighter shades I've measured a few that look to meat the next higher shade at their maximum transmission.

If you read the data sheet on speedglass you linked to you'll see its a variable intensity type which is shade 3 in its lighter state (consistent with the spectrum shown) and going up to shade 8-12 in it's dark state.

 

[archaeopteryx]

Hi Mike, thanks, adjustments made above. Seems to me the main thing is some specificity is desirable as to which welding protection might be repurposable as solar NDs. I've not had much luck finding shade 12-14 transmission curves which, for some years, I've been presuming is due to the spectrometer limits you mention. So, if you could link some of the darker shades you've found curves for, that would be interesting. Chou 2008 had a look with a Cary 5 and reported mean transmittance across a few bands for a range of "filters". The shade 12 and shade 14 included there were 14.3 stops in the IR versus 15.5 and 18.8 in the visible, which was rather a good showing.

Whilst I would hope the definition of continual viewing used was a more practical one, I'm unsure it's particularly worth guessing about the details. A more significant distinction is perhaps that optical instruments can be on the sun for the duration of an event, particularly if a tracking mount is used. For example, one camera I set for the 2017 eclipse had the sun continually in frame for about three hours.

It's marked MRC 16L

That just means it has sixteen layers, presumably eight on one side and eight on the other, with the outer ones having some resistance to being rubbed off. The resistance bit has been normal for decades and indicates MgF₂ or some other durable layer. I'm not aware of any test method for solar suitability less expensive than purchasing an ND100000. Mike is likely better to ask but it's my understanding fully suitable spectrometers are in the tens of thousands. So renting one probably costs hundreds. Perhaps if you know the right person, but that's a special case. If you want to test only to 1000-1100 nm those can be had for thousands but, at least where I am, rentals are three times full retail price for a new, higher end ND100000.

For IR ND evaluation the most accessible method I know is to look for color shift with a hot mirror (a UV+IR cut filter) added to the stack. This requires somewhat complex control for the color shift of the hot mirror as most (all?) are birefringent, B+W is the only manufacturer I know providing transmission curves for their hot mirror, and there's already a hot mirror in the sensor stack which provides something of a confound. So the test will indicate high near IR transmission but, if that's the case, probably the reason for testing the filter is it's casting too brown. In which case the test provides some insight as to why one doesn't like the filter but doesn't change anything about wanting to get a different filter. Because of the camera's hot mirror the approach doesn't provide much information for wavelengths longer than oh, 900 nm, or so.

If an IR or full spectrum converted body is available then more information can be obtained from test images using IR filters with different cutoffs. IR filters cost as much or more than NDs and several NDs add up to the cost of a conversion. So the reason to do this is really because one's curious of the details. I do wish this or spectrometry were common practice in ND filter reviews but it's understandable that they're not.

 

[Petrochemist]

Hi Mike, thanks, adjustments made above. Seems to me the main thing is some specificity is desirable as to which welding protection might be repurposable as solar NDs. I've not had much luck finding shade 12-14 transmission curves which, for some years, I've been presuming is due to the spectrometer limits you mention. So, if you could link some of the darker shades you've found curves for, that would be interesting. Chou 2008 had a look with a Cary 5 and reported mean transmittance across a few bands for a range of "filters". The shade 12 and shade 14 included there were 14.3 stops in the IR versus 15.5 and 18.8 in the visible, which was rather a good showing.

Whilst I would hope the definition of continual viewing used was a more practical one, I'm unsure it's particularly worth guessing about the details. A more significant distinction is perhaps that optical instruments can be on the sun for the duration of an event, particularly if a tracking mount is used. For example, one camera I set for the 2017 eclipse had the sun continually in frame for about three hours.

That just means it has sixteen layers, presumably eight on one side and eight on the other, with the outer ones having some resistance to being rubbed off. The resistance bit has been normal for decades and indicates MgF₂ or some other durable layer. I'm not aware of any test method for solar suitability less expensive than purchasing an ND100000. Mike is likely better to ask but it's my understanding fully suitable spectrometers are in the tens of thousands. So renting one probably costs hundreds. Perhaps if you know the right person, but that's a special case. If you want to test only to 1000-1100 nm those can be had for thousands but, at least where I am, rentals are three times full retail price for a new, higher end ND100000.

For IR ND evaluation the most accessible method I know is to look for color shift with a hot mirror (a UV+IR cut filter) added to the stack. This requires somewhat complex control for the color shift of the hot mirror as most (all?) are birefringent, B+W is the only manufacturer I know providing transmission curves for their hot mirror, and there's already a hot mirror in the sensor stack which provides something of a confound. So the test will indicate high near IR transmission but, if that's the case, probably the reason for testing the filter is it's casting too brown. In which case the test provides some insight as to why one doesn't like the filter but doesn't change anything about wanting to get a different filter. Because of the camera's hot mirror the approach doesn't provide much information for wavelengths longer than oh, 900 nm, or so.

If an IR or full spectrum converted body is available then more information can be obtained from test images using IR filters with different cutoffs. IR filters cost as much or more than NDs and several NDs add up to the cost of a conversion. So the reason to do this is really because one's curious of the details. I do wish this or spectrometry were common practice in ND filter reviews but it's understandable that they're not.

I'm afraid I've not kept links to any welding glass spectra I've found on line. I think I saved a JPG from one site which was at least reasonably informative, but finding it would take ages.

I'm not familiar with the Cary 5 it's from 1990 so predates my looking at replacement instruments & googles not finding significant details on it for me. I expect long integration times, wide splits & a lot of data smoothing has been needed to get anything from the dark shades. I'm sure Professor Chou has more experience in dealing with ultra dark filters than I have.

The solar viewing filters are sold in 2 types, one for direct viewing & a weaker one ONLY for photography. Instruments can often cope with radiation that can damage the eye & they are all more easily replaced.

You're spot on for the costs of spectrometers, our previous model (PE Lambda 35) cost £9k back in 2007. More basic models with less sensitivity can be found for as low as £2k, but many of those have too small a sample compartment to be of any use.

In the past I've found a good selection of hot mirror spectra at www.kolarivision.com nothing very up to date perhaps but including examples from many of the cameras they've converted. There is also spectral information available from the actual glass suppliers I've grabbed quite a few from UQG where I get my Schott glass.

It's perhaps worth pointing out that not all IR filters have sharp cutoffs. The value quoted is approximately the 50% transmission point (one of my filters was more than 20nm out on this, but most are nearer) In one example I have measured transmittance to 150nm below it's listed transition point. Here's a typical set of IR filters including that gradual change one (a 950nm in blue) & a section of processed negative:

Subscribe to see EXIF info for this image (if available)

 

[archaeopteryx]

I'm not familiar with the Cary 5

I couldn't find anything either but in looking around a bit more later I found Chou 2012 for the 2012 transit of Venus which graphs some of the data. It's squiggly and looks to maybe have some resonance issues.

Chou 2012 also provides retinal danger thresholds as a function of wavelength. 13.3 stops in the visible, 10 stops from 800-1400 nm and 0 at longer wavelengths. I'm not comfortable with those as there's no corresponding decrease in solar intensity at 800 nm and the 1400 nm threshold relies on energy absorption in the forward part of the eye. Personally, I would want something more like this as a minimum (constant transmitted spectral power density of 100 μW / (sr m² nm) using the 5250 K blackbody approximation) and prefer a couple stops darker. Clark 2017 performs a more detailed analysis and at arrives at similar results.

Subscribe to see EXIF info for this image (if available)

The solar viewing filters are sold in 2 types, one for direct viewing & a weaker one ONLY for photography.

I think you're referring Baader's AstroSolar 3.8 (12.6 stop) and 5.0 films. Other manufacturers differ. I would suggest 15+ stops. My solar exposures with a 15 stop ND32800 would be around f/5.6 1/6400 ISO 200. 12.6 stops would require around 1/32000 and I'm not aware of any electronic shutters that fast. Given most every camera and lens manual has some fine print about not pointing into the sun I'd be hesitant about leaving a mirrorless body or DSLR in live view for the duration of a solar event without 15+ stops as well.

It's perhaps worth pointing out that not all IR filters have sharp cutoffs

It's an inexact method to be sure. A more specific approach would be LED banks at various wavelengths, probably with a full spectrum converted body handing off to a SWIR camera around 1000 nm. That might come in under a thousand pounds, which is hundreds more than I could allocate to such a project, but at least it provides a measurement system with a longer lifetime than a spectrometer rental. Even with a converging lens I would anticipate some difficulty with long exposures and reciprocity failure at higher optical densities.

Also looked around some and there are welding glass patents into the 1970s, after which attention shifts to auto-darkening and twisted nematic properties (along with emission spectra of new welding methods as they emerge). That might be a contributing factor in the lack of welding glass spectra. What seem to be the end stage patents deal with oxide concentrations in the filter glass and some of the patents show high IR transmission. This makes me suspect a connection between welding filters and the, um, colourful oxide based ND filters which seem to be on their way out of the photographic market due to poor neutrality.

 

[Petrochemist]

The harm from NIR is as far as I know similar to that from visual light except that the iris of the eye does not close down to reduce intensity. I think that's why a greater degree of attenuation is felt necessary.

Yes it's the astrosolar film I'd seen. This is intended to be used via a telescope which will often darken things considerably more than f/5,6. There's certainly no need for such high ISO - I don't think the camera I used for snapping the sun through shade 13 welding glass could go beyond ISO 6400 IIRC I used 200. There is at least one MFT camera that can reach 1/32000 in electronic shutter. The damage to a camera from pointing at the sun will depend on focal length & time of exposure. There's a shot on the lens rental blog showing damage (a half melted shutter?) from use during the eclipse. It's not uncommon for landscapes etc to have the sun visible in the sky & these certainly aren't normally shot through 15 stop filters.

The SWIR (indeed anything over ~1100nm) is fairly irrelevant for photographic IR filters. Silicon as used in camera sensors becomes transparent at wavelengths above 1150nm, so the sensor does not respond to this at all. For similar reasons 1100nm is the upper limit to most UV/vis/NIR spectrometers (200nm is typically the bottom end as air becomes opaque below 190nm).

I've been meaning to try to set up a light source wavelength rig using my converted camera & a diffraction grating (a CD or DVD should do) something similar is shown on-line by public labs for producing spectra of filters. The cost once you have the converted camera would be minimal but calibration might be tedious. I have a more accurate spectrometer available, but it can't tell me the spectral power distribution of my light sources.
A bank of LEDs should work to give some indication of filter transmission but I suspect the LED wavelengths are no more accurate than the filters and as you say it probably wouldn't be a cheap set-up. Fortunately most IR filters have a reasonably sharp cut & the difference to images is generally not noticeable.
I've not been overly concerned with the spectra of very dark materials, when my standard method for filters proved inadequate for welding glass I tried a variant with slow scan speed & a wider slit, but didn't continue once that made little difference.

There may be some applications where a very narrow wavelength range is wanted perhaps to pick out an ink the blocks 778nm as opposed to one that blocks 805nm but this would be far from the realm of IR shot even by freaks like me I do have one filter with a very narrow (4nm) bandpass but I've not found a use for it (it came with a CCTV lens I brought)

I've yet to find any neutral density filters that are totally neutral. Even ultra expensive technical ones have ripples that can be seen in the manufacturers own spectra. Yes many of these are thousands of times more neutral than my ND filters, but for photographic use I can't justify even 1% of the cost of a 1cm wide filter - and there can't be many lenses where that would be wide enough!

 

[barry]

The damage to a camera from pointing at the sun will depend on focal length & time of exposure

Hi, this is a bit ambiguous, at least for Mirrorless.

I've heard of IR from sunlight melting components inside telephoto lenses as well.

 

[Bushboy]

My iPhone has a camera lens that could use a 1cm wide nd filter... wink wink ?
Seriously, you guys really know your stuff.
I’m gonna dig out my welding filters and have a crack at a pic with them.