Does anyone have any information on running a low pressure sodium light on a high frequency ballasts, I was thinking about building a ballast to run a lower wattage 18 to 35 watt lamp, and was wondering if the lamp will suffer from being driven at more then 50/60 Hz or if there is a maximum safe frequency (wouldn't normally be more then a few hundred kHz) and if the lamp prefers a square / sine wave.

The goal is to create a simple LPS ballast that is low cost, using easy to source parts and powered from 12 VDC (9 to 16) and run with an efficiency above 85% ideally more then 90%.

If anyone has any schematics they are willing to share that would also be greatly appreciated.   

The frequency is of no problem at all, there are even quite a few branded ballasts actually using HF electronic concept even for mains.
The problem will be the sensitivity of the lamps on the actual power they get (usually quite tight thermal budget in the lamp designs, if they are supposed to maintain their efficacy). And that would be the harder part for the required voltage range (the 9..15V is in no way excessive requirement fromthe battery perspective).

I see two options:Either you use a simple Royerish oscillator style (OCV and currenr according to the lamp specs) and use a separate switching pre regulator to keep the Royer input voltage in tight tolerances (e.g. at 8 or 16V using buck resp boost preregulator), or somehow tweak some "dimming" fluorescent ballast design so its current regulation feedback loop will mainly cover the supply voltage variation. But here the problematic part uses to be the startup - the com on IC's are designed for mains,so usually feature rather high startup unervoltage lock out threshold (after start it iseasy toderivethe elevated supply fromthe circuit itself, but the IC has to start first).
Or you may use somemore generic IC (IRS2153,...) with lower UVLO, butyou have to build around the required FM current control and protection from some discretes.

Or program the ballast control functionality into some microcontroller (e.g. ATTinyX5 family is suited for such applications), add power MOSFET gate drivers anddo it that way (it is again the same push-pull inverter plus L ballast with resonating cap).
I would  probably use the last way, as it allows the greatest flexibility (timing and setting of the supervision features) and promisses least complicated hardware, while still maintaining the efficiency.

How about step up power supply (to put out ~200V HF not stabilized), rectifier & capacitor, and standard "100..240V" fluorescent instant start ballast

@ Ash, energy budget is to tight I only have about 20 to 24 watts tops and the size of that is much larger then a dedicated power supply, I'm figuring I can build it all on a 2 inch by 2 inch double sided PCB with minimal heat-sinking if done correctly.

@ Medvac, I never though about using a micro-controller on it but that would drop the component count on the board and I can have it detect end of life as well allowing for an orderly shut down and if I go with a slightly higher memory I can probably have it control a timer as well, and add a gps so it turns on with sunset and sunrise, About the tight thermal budget as long as the lamp gets it's 18 watts it should warm up and operate and allow for one or two more watts as they age right ? And what sort of starting voltage due these tubes need, I was going to just use a simple strobe tube ignition transformer for this and superimpose it on the lamp input so it's separate from the main power supply.

What about just using a simple DC to DC and inverting the polarity every couple of seconds or several times a second would that cause any issues with the lamp, I know they don't like straight DC because of sodium migration?     

@ Medvac,

I'm Medved, not "Medvac" (I even don't know, what that is)... ;-)

I never though about using a micro-controller on it but that would drop the component count on the board and I can have it detect end of life as well allowing for an orderly shut down and if I go with a slightly higher memory I can probably have it control a timer as well, and add a gps so it turns on with sunset and sunrise,

Don't dream too much...
For such a low level ballast control, you need pretty fast and what is way more important, very accurate and predictable response from the SW. So really the supervision state machine (doing the startup/operation/shut down/failure mode and operation via an "LampPowerOn" input and "Fault" output interface pins) is the most complex thing you would be able to ever schedule into the operation loop beside the current measure-battery measure-frequency adjust loop.
For such extra features you would need a second processor, which will do all the bells and whistles of the high level functionality, while generating just that single "PowerOn" signal for the ballast controller processor...
The Tinyx5 have rather fast ADC, have an on-chip PLL allowing the peripherals to run on high frequency (if I remember well, it is in the order of 64MHz for peripherals and 16MHz for the CPU) still without an external crystal (never use anything else than fully internal RC oscillators for anything that should be at least somehow reliable)

About the tight thermal budget as long as the lamp gets it's 18 watts it should warm up and operate and allow for one or two more watts as they age right ? And what sort of starting voltage due these tubes need, I was going to just use a simple strobe tube ignition transformer for this and superimpose it on the lamp input so it's separate from the main power supply.

By the "tight thermal budget" I meant it needs much better than the 1:2 ratio for the full battery voltage range coming from otherwise very well behaving Royer oscillator without the preregulator.
The lamp actually needs the constant rated current (0.36A for the 18..35W lamp ratings), therefore the current regulation.
The required OCV is in the range of 300V for a fast glow to arc transition, plus about 500Vpeak for an ignition. Both are quite easy achievable using the resonant LC output.
As written before, Royer will do the job well, only needs constant accurate supply for that, so an upstream voltage preregulator. And that would make it less efficient than the frequency controlled resonant boost concept (step up plus the same LC as found in CFL's, only without the heaters or so).
A strobe tube ignition transformer is not designed to carry any arc current load, so it will burn right away. It is supposed to supply just pulses onto an external capacitive electrode, nothing more. So as the main arctube is behind the shield (it is conductive, but you can not electrically access it), the strobe transformer is useless for SOX.

What about just using a simple DC to DC and inverting the polarity every couple of seconds or several times a second would that cause any issues with the lamp, I know they don't like straight DC because of sodium migration?     

That may work, but I doubt it could be any less complex than any of the regulated AC concepts. Plus being a low pressure discharge, with HF drive you get few percents extra light output for the same power input (eddy currents within the discharge mass are pulling the current from the center towards the surface of the column, so shorter path for the generated radiation through the selfabsorbing environment, so more of it escapes the tube)...

On a mains supply I have had good success with a fulham WH3 for both 18 and 35w SOX lamps. I don't have any experience with DC ballasts, but this site looks to have some:

http://www.spgsolarcomponents.com/solarcomponents_en/solar-light-bulbs-lamps-tubes/low-pressure-sodium-lps/low-pressure-sodium-lps-e-ballast

Some tests I've done with 18w lamps: http://www.lighting-gallery.net/gallery/thumbnails.php?album=3893

I've tested a few more ballasts but never got around uploading the pics I've taken,.. maybe later.

SOX lamps actually run considerably better on HF operation than at 50/60Hz, in fact during the 1990s some manufacturers developed special SOX-HF lamps and matching electronic ballasts, but due to the higher cost they were not a commercial success.

Square wave operation is by far the best.  The efficacy of an LPS lamp is always higher when the discharge current density is lower, and at higher current the efficacy falls off especially quickly.  With a sine wave the peak current is much higher than RMS current and this momentary overloading limits lamp efficacy.  However with a squarewave drive it's possible to eliminate this peak, with a notable increase in efficacy.

Lamp lifetime can also be extended due to the possibility for softer starting.

James, where those HF lamps the SOX Pro lamps from philips ? I've read some where that you can get almost a 40% increase in system efficiency, but the major issue I've had so far is finding some information on the actual electrical characteristics of the lamp under different frequencies, and what the manufacturer claims to be ideal frequencies and waveform, because making a small square-wave power supply is not that complex, but it would be great to know the basic parameters, because depending on the frequencies one could drive a 180 watt SOX from something as simple as a small automotive inverter with some minor modification and do this for under $20.     

The ballast output should behave as a constant current source with high output impedance. Plus the high impedance has to be maintained till really high ftequency (from the ionization inertia perspective). That means whatever you make there, it should end by an inductor or very limited capacitance (max 10nF or so).
With these limitations the real square wave output ballast become quite extremely complicated, definitely not any common power supply or so. Thecomplexity brings extra losses as well and I don't think the extra efficacy you may get is worth the complexity and the risk of loosing it back in the form of the extra ballast losses (unlike big company radying for mass production, you won'thave the budget to really finetuneal the details).
So way easier is to make the ballast really high frequency resonant type. There you  ay not reach the efficacy of the square wave, but still it will bebetter thanthe 50/60Hz and it will definitely be a way simpler and less lossy (mainly when taking into account the limitations of home brew development) circuit...
Don't forget the claimed 40% or more improvement on the system efficacy goes mainly on account of eliminating the huge losses of magnetic SOX gear (the autoleaks have efficieny in the range of 60% or even less, the series choke is around 70%, while with even a dodgy electronis you are easily around 90% energy efficiency) the differences in lamp efficacy alone are in the range of 10% or so, so something I don't see worth complicating the ballast design and risking the extra ballast losses eating all that gain back.

Would this work :

HF current source at freq F1 ----> rectifier ----> H bridge making squarewave at freq F2

F1 >> F2

The lamp sees pretty much F2 squarewave

Medved, how high of a frequency would you suggest ? a couple of kHz or a few MHz, And do you think there is any risk from the acoustic streaming effect inside the arc tube when using high frequency and causing instability in the lamp ? I know small MH (39 watt) lamps don't like much over 40 kHz, otherwise they would excel at about 300 kHz if it wasn't because of this.. For the actual lamp start up I was thinking of using an ignition pulse superimposed on the output, but I'll stay away from those strobe tube pulse transformers as you pointed out some critical drawbacks I over looked and I guess you only need a few hundred volts anyhow so that shouldn't be hard to generate, so the rest of the circuit can be designed to just drive the lamp and let the igniter do the starting, and what do you think about borrowing from a simple automotive type inverter design, and switching at a high frequency to step up the voltage, rectifying it and filtering it, then using a couple of fets to switch it again so we are playing with nice clean DC and allowing for better control of the output, I know this will lose a few watts by double switching it but then I should be able to better control the tubes input and make the lamp last longer and I could also then get a much wider input voltage range without much effort ..  

Acoustic resonances, aren't problem with LPS lamps, as they are low pressure and low temperature discharge lamps, similar to fluorescent lamps.
They can only damage HID lamps, such as HPS and MH lamps.

It is a low pressure lamp (the way it operates it even does not belong to the HID family), so no problems there at all, it is the same as fluorescent (the same discharge type; just mercury instead ofsodium).
So from efficiency perspective (include circuit parasitic effects) and to prevent noise (both direct accustic, as well as disturbing the 19kHz FM stereo pilot signal) the best would be something between 20 till 50kHz (over the battery voltage range, as the frequencywill have to vary to compensate).
I would use an IRS2153 driving the primary push pull FETs, but the "timing" capacitor is not connected to GND, but to the "PWM" output of an ATTiny25. The IRS RC is set just below the 20kHz, the ATTiny then synchronize it to the frequency it generates. Note the generatedduty ratio should be all the time exactly 50% (the code should change both period, as well as the threshold at once and keep their relation exact), otherwise the core will saturate and blow the transistors or the input DC filter capacitor.

@Ash: You forgot the filter (without that no regukation by PWM is possible at all. And there lies the problem: The filterneeds to have an inductive input, bean LC, what means capacitive output. And the capacitance is not liked by the discharge, so it should be very limited.
Plus that concept means three stages: Inverter-rectifier-inverter. Each step brings its losses (the rectifier induced losses don't end up in the dioes, but within the first inverter). So try to compete on efficiency with just a single stage, moreover operating in low loss soft switching mode...