Gakken is a name familiar to me, as it was their "Denshi Block" SR3A electronics kit given to me in 1976 which did a lot for my electronics knowledge. It was with this kit I first constructed simple transistor amplifiers and radios, with a few novelty circuits as well. This kit was the predecessor to the more well known EX150 of more recent times.
In recent years they seem to have re emerged with a line of very interesting kits, some electronic, and some mechanical. Readers of this site would be mostly interested in the little valve amplifier, the radio kits, and the acoustic gramophones. These are well known to those who trawl eBay, and there's a lot of sites related to them - mainly describing what the author thought of constructing them.
There's little technical information however, so this article will make up for that. First, put "gakken vacuum tube radio" into your search. It is worthwhile searching the Japanese sites via http://www.google.co.jp, seeing as they are of Japanese origin. I've built both the one valve variometer tuned set, and the three valve regenerative set. Apparently, the valves used are old stock Chinese valves. They have unusual type numbers, but are directly equivalent to the usual RMA and European 7 pin battery valves. HT supply in both kits is from series connected 9V batteries, while the heater supply is a single C cell.

Variometer Tuned Receiver.

I bought this kit, late February, 2011. The way it was presented is certainly very Japanese, and for US$39 was excellent value. The receiver comes in a box dimensioned like that of a large book, and the instructions come in a bound magazine. It's a very professional job and the detail gone into is quite considerable. I have never seen a set of kit instructions so detailed with regards to how the receiver works. It goes into some detail about the variometer tuning, even providing mathematical formulas. Perhaps it's because the Japanese aren't into being dumbed down like the rest of the world. Most kit instructions I've seen tell very little of the circuit design.
The construction of this kit is very easy. All it basically entails is winding the two coils and connecting them to the PCB under the plastic chassis. There is some minor assembly putting in a few self tapping screws. I was mainly interested in the technical aspects however, which is why I bought it, and here I was quite surprised at what I found.
I had assumed, after building the three valve set a few years ago, that this set would use a pentode (1T4 or equivalent) as a regenerative detector. Not so!

The circuit.

The tuned circuit feeds a reflexed RF/AF amplifier without regeneration.

It's a reflex circuit without regeneration, which is a very Japanese thing. Look at any simple transistor radio, or electronics kit designs, from Japan in the 1960's and you'll see reflex circuits everywhere.
They are the basis of those two transistor "Boys Radio" designs popular in the U.S. Performance with them is generally good, but by adding regeneration you'll get something very similar to superhet performance. Without regeneration, they are only good for local reception. The idea is to amplify the incoming RF, detect it, then send the resultant audio back into the valve to be amplified further. The one valve or transistor functions both as an RF and AF amplifier.
First, let's look at the tuned circuit. With the two coils in series it is possible to change the inductance by swinging one through 90 degrees to the other. Inductance decreases with the coils further apart. This inductance in conjunction with the variable capacitor (according to the book it's 100pF) and optional 100pF is what determines the tuned frequency.
As the range of inductance is insufficient to cover the whole broadcast band, an extra 100pF can be switched across the coil to receive stations at the low frequency end.
With the coils 90 degrees to each other, the 100pF switched out, and the variable capacitor at minimum it tunes to about 1600Kc/s. With the coils parallel and maximum capacitance, it tunes to about 550Kc/s.

Reflexing.
This is the process of using one amplifier for both RF and AF. As the two bands of frequencies are so far apart we can amplify both with the one valve, without any significant reaction between the two. Incoming RF is fed into the grid with an amplified signal present at the plate; it's developed across the RF choke. The amplified RF is fed via a 100pF to a voltage doubling detector using germanium diodes. The diode load and RF filter is formed by the 1M and 220pF. At this point the audio signal becomes available. So far, the description applies to any RF stage feeding a detector, and indeed a simple receiver can be made thus. However, we can 're use' the amplifier valve for the audio signal as well. So, into the grid it goes via the 330K isolating resistor. This resistor prevents the incoming RF from the aerial coil being shunted to earth by the detector circuit, and in particular the 220pF. Yet, it allows the audio to flow into the grid unimpeded. The RF choke has little effect at audio, so can be regarded as a short circuit. However, the audio choke has a high impedance and it's this which allows the amplified audio to appear at the plate. The signal is fed to a piezoelectric earphone, via a .1uF and 1M to eliminate any DC being applied.
A red led shows HT current is flowing and acts as a power on indicator (the valve filament is not really visible). The heater current comes from a C cell in a bayonet fit cylindrical holder above the chassis - power switching is achieved by rotating the enclosure so the contacts touch or break as required.

Pentagrid valve.
The most bizarre thing which I didn't notice at first, is they've used a pentagrid valve wired as a pentode. It's a 1A2 which is equivalent to 1R5. I can only guess this was done because they had lots of pentagrids left over (the other three valve kit doesn't use them), or because one of the experiments involves converting the set to a superhet. In fact, the book goes into quite lengthy detail about modifications to the basic circuit. The only trouble is I can't read Japanese! The superhet involves installing a couple of transistor type IF transformers which then feed an LMF501 (ZN414) IC for further amplification and detection. There is also a modification to convert the unit into an RF oscillator used for Morse code transmissions. Most of the information is centred on the direction finding qualities of the loop aerial, and one very large wood framed aerial is shown.

Performance.
Unfortunately, this was quite a let down. With no regeneration, it is only possible to receive 2FC and 2BL at my location. Volume is similar to a crystal set. There appeared to be an intermittent fault with the tuning capacitor; it certainly doesn't look like it's made for continuous use. Also, there was at times instability and breaking into oscillation.
I did try coupling my outside aerial into the tuned circuit, both inductively and capacitively. Unfortunately there was very little improvement, although with inductive coupling I could receive some interstate stations in the evening. I suspected there may be some underlying fault because I expected to at least be able to receive all the Sydney stations with no external aerial.
The use of the pentagrid had me suspicious and I sensed that used as an RF amplifier the gain migh be a bit low. It would seem to me that signal fed into the 3rd grid would result in less gain. So, I thought about a pentode. As it happens, a IT4 can be plugged into the valve socket with no modification. The plate and heater pins are the same. Pin 4 on the IT4 is not used. Pin 3 is the screen grid of the 1T4, and pin 6 is the control grid. The results were much improved, and were as they should be. The intermittent instability was also now gone.

This was the first of my Gakken radio kits; I purchased it in August 2006. It has been superseded with a "Version 2" which includes a piezoelectric microphone for testing the audio stages, and a larger output transformer. It was largely the 1920's look alike with the frame aerial and horn speaker that attracted me to it.
This kit has been around longer than the variometer tuned set so there's a bit more on the internet about it. Again, it was only really intended for the Japanese market but the pictorial quality of the manual makes for an easy assembly even if you can't read the words. One important point is the variable capacitor. Peel off the plastic film from one of the plates only. The film remaining on the other plate functions as dielectric. Since purchasing my kit, an English manual has appeared. It can be found here.
 
 

Gakken Vacuum Tube Radio kit is a standard regenerative design.

Regenerative Detector.
There is nothing unusual about the circuit, except they aren't using the valves to full potential. The 1B2 and 2P2 are wired as triodes whereas more gain, and output could be had with them in pentode configuration. The 1K2 functions as a pentode regenerative detector with the usual grid leak components, 270pF and 2.2M in conjunction with the diode formed by the grid and cathode (heater). Giving a slight positive voltage to the grid improves sensitivity which is why the gird resistor is returned to the positive heater terminal. Due to the use of a frame aerial, the receiver is self contained needing no external aerial. It does however mean there's a strong directional characteristic. An unusual feature is the so called 'book capacitor'. It simply involves moving two hinged plates closer or further apart to set the capacitance. The manual implies this was a common feature in the 1920's. I've seen a lot of 1920's receivers and not one yet has used this method of tuning.
Positive feedback is obtained with the smaller winding in the plate circuit. The degree of feedback is controlled by shunting a 10K pot across this winding, progressively loading it.
My main criticism is the regeneration control. Forcibly lowering the Q of the tuned circuit  by shunting it with a rheostat to control oscillation is rather crude. The pot setting is very critical and hard to optimally adjust. Backlash is also a problem. Once it has started oscillating you have to turn the pot further back than the point was where it starts to oscillate.

The PCB for Version 1 with wires connected. The instructions show the aerial coil wires just twisted to the plastic wires from the PCB, but I soldered mine. The C cell connections are visible to the right.

Version 2 PCB. This set has not been completed.  Note that pin 4 of the output valve (designated Yellow) is connected to the 45V supply for pentode operation.

Audio stages.
Detected audio present at the plate then proceeds to the next valve for further voltage amplification. This is a pentode wired as a triode. Using it this way results in less gain so I am curious as to why it has been designed this way. A much amplified audio signal is available at the 1B2 plate for feeding the piezoelectric earphone, or the output stage. Again, wiring the 2P2 as a triode reduces gain and power output. The speaker transformer is much smaller than would normally be used; it's about the size of something found in a small transistor radio. As for the speaker itself, this is merely a driver from a set of Walkman style headphones. Historically, that's not too far off the mark because the horn speakers of the 1920's did in fact use an oversize headphone driver.
The led indicates power on due to the HT current flow. The voltage drop across it and the 270R resistor also creates the negative bias for the 2P2. An earphone is supplied with the kit and can be plugged into the socket provided. Here, the earphone is fed from the 1B2 and when its plug is inserted, the normally closed contacts in the socket break the heater connection to the 2P2. This reduces the current drawn from the C cell by 100mA. HT supply is five series connected 9V batteries giving 45V which is lower than the output valve is actually rated for.

Performance.
In terms of RF sensitivity and selectivity, it's like any other regenerative set. Interstate stations are easily received at night. Apart from this, the radio has some serious limitations. These could be fixed at the expense of losing originality. The worst aspect is the tuning control. The 'book capacitor' suffers terribly from mechanical instability. Not only does the adjustment screw thread have slop in it, but just any pressure on the chassis distorts it, and the position of the two 'plates'. It is rather difficult to tune weak stations because of this and one has to gently prod the tuning control until the station comes into tune. Next on the list is the regeneration control. As I expected it would, it does suffer from backlash and critical adjustment. Also, the pot isn't perfect and simply touching the knob without even turning it will alter the resistance enough to throw it into oscillation. Thirdly, is the low audio gain.
Three valves should drive the output into overload, but with this set it's good enough for a quiet room but not much more. I did rewire the output stage to a pentode which improved things somewhat, but I think the audio voltage amplifier stage needs to be treated likewise. The tiny speaker transformer and headphone driver look dubious anyway. Nevertheless, I did try a full size speaker and transformer, and while there was an improvement it wasn't startling. Incidentally, I found troubles with with earphone socket and a dry joint - the radio would not always produce sound.

How to improve it.
First thing would be replace the 'book capacitor' with a proper tuning gang. Secondly, the regeneration control needs to be improved. Maybe a high quality pot would reduce the critical adjustment, but it wouldn't fix the backlash. The best way would be to control the screen voltage of the 1K2. The 10K pot would have to be replaced with something like 100 - 500K, used as a voltage divider across the 45V supply to feed the detector screen grid.  Quite a few PCB tracks would have to be cut and joined to accommodate the changes.

For the audio stages, higher output will be obtained if the output valve is wired as a pentode. As it happens, the Version 2 circuit has made this improvement.
A larger speaker transformer should be used, but to find one with a 10K primary that fits under the chassis is the awkward part here. And finally, to the headphone driver, a proper miniature speaker may provide an improvement. Jacking up the HT would help too, but the problem is where to fit the extra 9V batteries.

Pentode connection for the output valve. Note track cut at pin 4 and short blue wire added.

To connect the output valve as a pentode (Version 1), cut the track to pin 4, and then connect pin 4 to the 45V supply. Note the short blue wire in the photo above.
A further increase of gain would be obtained by connecting the 1B2 as a pentode. I have not yet tried this, but if you wish to experiment, cut the track between pin 4 and 5 so that pin 4 is completely isolated. Next, connect a 0.1uF capacitor from pin 4 to earth (you can use pin 7 for this). Finally, connect a resistor of 1M between pin 4 and the 45V supply. This resistor might not be an optimum value, so there is room for experimentation. I would expect best results to be somewhere between 390k and 2.2M.