Once I realized there was a schematic on your original file, I was able to view it with NeoOffice. Given the inefficient design you were provided, I would be inclined to start over with a switching-type circuit. That would be Part One of the project.
The price you were quoted for the LM3423 must be for the evaluation board, including all of the parts and the circuit board, plus import duty and other taxes. The chip itself is $4.15 US from http://www.digikey.com. As you noted, the PWM control input is not the easiest thing to use given your analog input signal.
I have another one for you to look at - the LM5088. This part costs less - $3.80 US. It is a general purpose bucking switch-mode regulator. Referring to the schematic on the front page of the manufacturer's datasheet (http://www.national.com/ds/LM/LM5088.pdf), one of your LED series strings would replace Rfb2, and its current sharing resistor would be Rfb1. The other strings would simply run from the output to ground. The current sharing resistors would be 1.205V/Imax, or 1.0 Ω at 1.205 Amps. You can control the output current up to the maximum current by applying your signal to the SS pin, provided that the signal is no more than 1.205V. If the control voltage waveform exceeds that level, your LEDs will simply run at the maximum current (a control voltage over 7 Volts will damage the regulator).
Note that the output transistor circuit has its own current sense resistor. This circuit is sized for the total output drive current. Also, the on-resistance of the transistor must be low enough so the source-drain voltage is less than 1 V at maximum current, otherwise the regulator's protection circuitry will shut down everything. Those are some particular items I picked up from a brief perusal of the data sheet.
The drawback of this design is that instead of pulsing the maximum set current to get the desired average, you are actually reducing the set current with the control voltage at the "soft-start" input. This may produce inconsistent color. The LM3423 and similar products are designed to get around this issue. It appears that what you would like to have is something like the LM3423 with an analog control input rather than a PWM input. You would be looking for something designed to drive an external transistor, and your input power supply voltage needs to be at least 40V. I've only looked at National Semiconductor parts; other manufacturers may have a better match. I like to use National Semiconductor parts as a starting point because they have some of the best (most detailed and informative) data sheets in the industry. I learned as much about electronic design from these as I did from my formal studies. Certain far east manufacturers would benefit greatly from their example...
The next question is the maximum frequency of the input control voltage. If the project is a light panel for special effects, I would not expect it to be much over 60 Hz, since the human eye does not readily discern flicker above that frequency, even in the relatively faster peripheral vision regions. Even animals with only monochrome vision would not notice flicker much above that frequency. That is why you can watch television with a 60 Hz vertical frequency - you don't perceive the individual frames or sweep lines; rather your eye and brain integrate the raster into a single time-continuous picture.
If I understand the application requirement correctly, then even at its slowest speed of 50 kHz, this regulator will have enough bandwidth for the task. Certainly at 1 MHz it should be able to keep up with variations below 20 kHz. 200 kHz is probably good enough, but you may want to check the radio spectrum usage in your region to make sure that the fundamental and first three harmonics won't interfere with anything.
Now for the question of a sine wave represented by PWM: if the PWM waveform is integrated, you will get the sine wave back again (plus possibly some harmonics and other results of errors in the process). The on-time fraction of each pulse is the ratio of the instantaneous input voltage to the reference voltage (assuming the input is less than the reference). Negative voltages won't work, so your waveforms will need a DC offset to avoid cut-off at the bottom, just as in your analog current source.
There are many ways to obtain this ratio. A relatively simple analog-based method is to sample the instantaneous voltage and compare it with a ramp waveform running at a frequency much greater than that of the waveform to be converted. When the ramp = sample voltage, the pulse is switched off, so it stays on longer for higher voltages. At the end of the ramp cycle, the input voltage is sampled again and the process is repeated. (Sampling: using a solid-state switch, connect the voltage to a small, very stable, low leakage capacitor. Open the sample switch, then do the ramp comparison with the voltage on the capacitor. See also "sample and hold amplifier.") For lesser precision and easier timing requirements, the input voltage is applied directly to the ramp comparator.
How are you coming with Part Two (thermal management)? The high-efficiency PWM circuit will get rid of most of the heat from the current source electronics, but you still have to deal with the thermal losses in the LEDs. The thermal power figure I calculated for the last post was worst case, assuming all of the input energy went directly to heating the LED. Fortunately, most of it should be going to light production instead. Even so, you have to keep the total thermal resistance to ambient as low as possible to avoid having to dim the light at high ambient temperature or frying the devices. See Figure 25 in the data sheet for the K2 LED - this gives you an idea of how much you would have to cut the current at high temperatures for various heat sink sizes (including the thermal resistance of the LED).
If you want automatic thermal protection for your LEDs, you could use a temperature sensor circuit to clamp or reduce the magnitude of the control voltage the temperature increases. The reduction rate would be set to match or exceed the appropriate derating curve in Fig. 25. The curve you can use depends on how effective your heat sink is. For example, if you have a good-sized heat sink with a thermal resistance of less than 2°C/W per LED (assuming you are taking the heat from star plates), you can use the top curve and will probably rarely if ever have the de-rating function kick in.
The tricky part with heat sink mounting is the connection to the LED string used for regulating the current. You want the top of the string to be close to the output inductor, and you want the bottom LED of the string to be close to the feedback /current sharing resistor, which must be close to the regulator IC. This string won't be a straight line since it would have to fold in the middle to get back to the regulator.
I recommend making sure your heat sink has a solid connection to the power ground at the regulator; this should prevent a lot of problems with noise and instability (it would be almost like a ground plane in a circuit board).